GW AI Fluency Wiki - User contributions [en]

Your First AI Team Meeting

2026-03-16T16:28:26Z

Admin: Fix 1 internal link(s)

''Run a multi-perspective AI session with two expert viewpoints on the same problem. 15 minutes.''

<blockquote>
'''One-liner:''' Run a multi-perspective AI session where one prompt gets you two expert viewpoints on the same problem — no extra tools required.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a real decision you're currently facing. It could be a work decision, a project direction, or a problem you're stuck on.

Paste this prompt into any AI chat (ChatGPT, Claude, Gemini — anything works):

<blockquote>
I want you to act as two different experts giving me advice on '''[your problem here]'''.

First, respond as a '''[Role A]''' — someone who focuses on '''[their priority]'''.
Then, respond as a '''[Role B]''' — someone who focuses on '''[their different priority]'''.

Keep each perspective clearly labeled. Be specific and give concrete recommendations, not vague advice.
</blockquote>

'''Example — choosing whether to launch a feature now or wait:'''

<blockquote>
I want you to act as two different experts giving me advice on whether to launch our new onboarding flow this week or wait until next month.

First, respond as a '''growth-focused product manager''' — someone who prioritizes user acquisition and speed to market.
Then, respond as a '''risk-aware QA lead''' — someone who prioritizes stability, edge cases, and user trust.

Keep each perspective clearly labeled. Be specific and give concrete recommendations, not vague advice.
</blockquote>

After reading both perspectives, send this follow-up:

<blockquote>
Now, act as a '''neutral facilitator'''. Summarize where these two experts agree, where they disagree, and what the key trade-off is. End with a single question I should answer before making my decision.
</blockquote>

Read the synthesis. Notice how one prompt gave you a structured debate that would normally require two people in a room.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose your problem''' — Pick a real decision or challenge you're working on right now. It works best when reasonable people could disagree about the right approach.
# '''Pick two expert roles''' — Choose two perspectives that would naturally see your problem differently. Examples: marketer vs. engineer, short-term thinker vs. long-term strategist, customer advocate vs. operations manager.
# '''Write and send the dual-role prompt''' — Use the template in the "Jump in" section. Fill in your problem and your two roles.
# '''Read both perspectives''' — Notice where they conflict, where they agree, and which one you instinctively lean toward.
# '''Send the facilitator follow-up''' — Ask the AI to synthesize the two views and surface the core trade-off.

'''"Done" looks like:''' You have a summary of two contrasting expert viewpoints and a clear understanding of the key trade-off in your decision.

----

== 🧭 Why this matters (Strategists start here) ==

This exercise builds the foundational skill behind all multi-agent AI workflows: '''defining specialized roles and comparing their outputs'''. At the intermediate level, you'll split these roles across separate AI sessions with different contexts. At the advanced level, you'll design entire agent architectures. But it all starts here — training yourself to think in terms of roles, perspectives, and structured disagreement rather than asking AI once and accepting the first answer.

----

== Reflection ==
* Did one perspective feel stronger than the other? Why — was it genuinely better argued, or did it just align with what you already believed?
* What did the facilitator synthesis surface that you hadn't considered?
* Would you use this dual-role technique for real decisions going forward? What types of decisions benefit most?
* 💬 ''Run this exercise with a colleague in the room. Have them choose different expert roles than you did for the same problem — the role selection itself reveals different priorities.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[The Handoff Protocol|AC-Intermediate-01]] — where you'll split these roles across separate AI sessions and learn to manage handoffs between them.

Back to [[Agent Collaboration|Agent Collaboration]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Agent Collaboration Exercises]]

Why AI Gets Things Wrong

2026-03-16T16:28:25Z

Admin: Fix 4 internal link(s)

''AI doesn't lie — it generates plausible text. Understanding why helps you catch mistakes before they matter.''

<blockquote>
'''Plain English:''' AI produces confident-sounding text that is sometimes completely wrong. This isn't a bug — it's a fundamental feature of how these systems work. Knowing why it happens is the single most important thing you can learn about working with AI.
</blockquote>

== What's actually happening ==

When AI writes something incorrect, people call it a "hallucination." The word is a bit misleading — it implies the AI is seeing things that aren't there, like a glitch. What's actually happening is simpler and more important to understand:

AI generates the most statistically likely next words based on patterns in its training data. It has no mechanism to check whether what it's producing is true. It doesn't "know" things the way you know your own phone number. It produces text that ''looks and sounds like'' correct text, because it learned from millions of examples of correct text.

This means:
* It can write a perfectly formatted citation for a paper that doesn't exist — because it's learned the ''pattern'' of what citations look like.
* It can confidently state a statistic that's completely fabricated — because it's generating a plausible number in a plausible context.
* It can describe a product feature that was never built — because the description sounds like something that ''could'' exist.

The AI isn't lying to you. Lying requires knowing the truth and choosing to say something different. AI doesn't know the truth. It's generating plausible text. That's a crucial distinction.

== When it's most likely to go wrong ==

Hallucinations aren't random. They follow predictable patterns:

'''Specific facts, numbers, and dates.''' Ask AI for a general explanation of how photosynthesis works and it'll be accurate. Ask it for the exact year a specific obscure paper was published and it might invent one. The more specific and verifiable the claim, the more you need to check it.

'''Citations and sources.''' AI is particularly bad at this. It will confidently produce author names, paper titles, journal names, and URLs that look real but don't exist. Never trust an AI-generated citation without verifying it.

'''Recent events.''' AI models have a training cutoff date. If you ask about something that happened after that date and the AI doesn't have search access, it may either say it doesn't know (good) or generate a plausible-sounding answer (dangerous).

'''Niche or specialized domains.''' AI performs best on topics that appeared frequently in its training data. Mainstream topics in English have dense coverage. Obscure or specialized topics — especially in other languages — have less, so the model has fewer patterns to draw from and is more likely to fill gaps with plausible-sounding fabrications.

'''When you push it.''' If you insist the AI answer a question it's uncertain about, or tell it "you must provide an answer," it will comply — by generating something. AI tools generally don't have a strong instinct to say "I don't know." Some are better than others, but the pressure to produce output is built into the system.

== Why it ''sounds'' so confident ==

This is the part that trips people up. When a human says something confidently, you assume they believe it and probably have some basis for it. When AI says something confidently, it means nothing — confidence is the default mode.

The model isn't more certain about accurate statements than inaccurate ones. It generates all text with the same fluent, authoritative tone because that's the pattern in its training data. Well-written text sounds confident. The model produces well-written text. Therefore, everything it produces sounds confident — including the wrong things.

This is why the philosopher Harry Frankfurt's essay "On Bullshit" is on our [[Further Reading|Further Reading]] page. Frankfurt distinguishes between lying (knowing the truth and hiding it) and bullshitting (not caring whether something is true). AI is, technically, the world's most sophisticated bullshit generator. Not because it's trying to deceive you, but because truth and falsehood aren't categories it operates in. It operates in plausibility.

== What you can do about it ==

The good news: once you understand this, you can work with it effectively. The [[The Fact-Check Habit|Fact-Check Habit]] and [[The Verification Checklist|Verification Checklist]] exercises build these skills in practice. Here's the framework:

'''Treat AI output as a first draft, not a final answer.''' This shift in mindset is the most important thing. AI gives you a starting point — fast, broad, often useful. Your job is to validate, refine, and apply judgment.

'''Cross-reference specific claims.''' If the AI states a fact, a statistic, or a date that matters to your work, verify it with a primary source. This takes 30 seconds and prevents the kind of embarrassing errors that erode trust.

'''Watch for the patterns above.''' You now know when hallucinations are most likely. Apply extra scrutiny to specific facts, citations, recent events, and niche topics.

'''Ask the AI to flag its uncertainty.''' Adding "If you're not sure about something, say so" to your prompt doesn't guarantee honesty, but it does help some models hedge appropriately rather than fabricating with confidence.

'''Use AI features that ground responses in sources.''' Tools like Perplexity, Claude with web search, or ChatGPT with browsing can cite where they found information. This doesn't eliminate errors, but it gives you something to check against.

'''Don't ask AI to be your only source.''' AI is best when it's one input among several — when you're using it alongside your own expertise, your colleagues' perspectives, and verified data. The [[The Multi-Source Brief|Multi-Source Brief]] exercise practices exactly this.

== The bigger picture ==

Understanding hallucinations isn't just about catching mistakes. It fundamentally shapes how you think about AI's role in your work:
* It's why the [[Pillars/Ethical Prompting|Ethical Prompting & Judgment]] pillar exists — because using AI responsibly requires knowing its limitations.
* It's why human oversight matters in any AI workflow — and why the most advanced exercises in this playbook always include human checkpoints.
* It's why AI fluency is more than just "knowing how to prompt" — it's knowing when to trust, when to verify, and when to override.

== Where to go next ==
* [[The Fact-Check Habit|The Fact-Check Habit]] — practice catching AI mistakes in 15 minutes
* [[Prompt Engineering Basics|Prompt Engineering Basics]] — techniques that reduce (but never eliminate) hallucinations

[[Category:AI Fluency Playbook]]
[[Category:Core Concepts]]

Tokenization & Context Windows

2026-03-16T16:28:23Z

Admin: Fix 3 internal link(s)

''Why AI can only 'remember' so much, why long conversations go off the rails, and what you can do about it.''

<blockquote>
'''Plain English:''' AI doesn't read words — it reads "tokens" (word chunks). Every AI tool has a limit on how many tokens it can handle at once. This is the context window, and it's the single biggest constraint on what AI can do for you.
</blockquote>

== What tokens actually are ==

When you type a message to AI, it doesn't see words the way you do. It breaks your text into smaller pieces called '''tokens'''. Sometimes a token is a whole word. Sometimes it's a piece of a word. Sometimes it's just punctuation.

For example:
* "Hello" → 1 token
* "Tokenization" → might become "Token" + "ization" → 2 tokens
* "I'm building a workflow" → roughly 5 tokens

A rough rule of thumb: '''1 token ≈ ¾ of a word''' in English. So 1,000 words is roughly 1,300 tokens. A 10-page document is roughly 4,000–5,000 tokens.

Why does this matter to you? Because every AI tool charges by tokens and limits by tokens. When you hit a message length limit, get a "conversation too long" error, or notice AI "forgetting" things you said earlier — that's all about tokens.

== The context window: AI's working memory ==

The '''context window''' is the total number of tokens the AI can process at once. Think of it as a whiteboard: everything in your conversation — your messages, the AI's responses, any documents you've uploaded, the system prompt — all has to fit on this whiteboard. When it's full, things start falling off the other end.

Current context window sizes (as of early 2026):

| Model || Context window || Roughly equivalent to

| Claude (Anthropic) || 200K tokens || ~150,000 words — a full novel
| GPT-4o (OpenAI) || 128K tokens || ~96,000 words
| Gemini 1.5 Pro (Google) || 1M+ tokens || ~750,000 words — multiple books

These numbers sound enormous, but they fill up faster than you'd think. A long conversation with back-and-forth responses, a few uploaded documents, and a detailed system prompt can eat through 200K tokens in a working session.

== Why this matters for your work ==

'''Long conversations degrade.''' If you've noticed AI giving worse answers later in a conversation than at the beginning, this is why. As the context window fills up, the model has more text to process and older information gets less "attention." Starting a fresh conversation for a new topic isn't a sign of failure — it's good practice.

'''Uploaded documents have limits.''' When you upload a PDF or paste a long document, it consumes context window space. A 50-page report might use 20,000+ tokens, leaving less room for your actual questions and the AI's responses. If you're working with long documents, consider summarizing or extracting the relevant sections first.

'''"AI forgot what I said" is usually a context issue.''' AI doesn't have memory between conversations (unless you're using features like Claude's Projects or custom GPTs that provide persistent context). Even within a conversation, if you're 30 messages in, the AI may lose track of something you said at the beginning because it's being pushed out of the active window.

'''This is why the [[The Handoff Protocol|Handoff Protocol]] exercise matters.''' When you learn to structure handoffs between AI sessions — summarizing context, carrying forward the essential information — you're working around context window limits intelligently.

== Practical tips ==

'''Start fresh for new topics.''' Don't keep one mega-conversation running for everything. A new topic deserves a new conversation with focused context.

'''Front-load the important stuff.''' Put your most critical instructions, context, and constraints at the beginning of your prompt. Information at the start and end of the context window gets more "attention" than information buried in the middle.

'''Summarize before you continue.''' If a conversation is getting long and you want to keep going, ask the AI to summarize the key decisions and context so far, then start a new conversation with that summary.

'''Use persistent context features.''' Claude Projects, custom GPTs, and system prompts let you set context that persists across messages without eating into your per-message token budget. The [[The Handoff Protocol|Handoff Protocol]] exercise teaches you to design these.

'''Be selective with document uploads.''' Instead of uploading a 100-page document and asking a question, extract the 5 relevant pages. You'll get better answers and use less of your context budget.

== Where to go next ==
* [[Prompt Engineering Basics|Prompt Engineering Basics]] — how to make the most of limited context
* [[The Handoff Protocol|The Handoff Protocol]] — practice managing context across AI sessions

[[Category:AI Fluency Playbook]]
[[Category:Core Concepts]]

The Workflow Blueprint

2026-03-16T16:28:23Z

Admin: Fix 2 internal link(s)

''Design, document, and test a complete AI-automated workflow for a real business process. 40 minutes.''

<blockquote>
'''One-liner:''' Design, document, and test a complete AI-automated workflow for a real business process — from trigger to output, with error handling and quality gates.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a real business process that currently takes you 30+ minutes and involves multiple steps. Examples: weekly reporting, content production, customer onboarding documentation, project status updates, invoice processing.

'''Phase 1 — Map the current process.''' Send this:

<blockquote>
I'm going to automate this business process: '''[describe the process]'''.

Help me map the current manual workflow:
1. What triggers the process? (time-based, event-based, request-based)
2. What are the sequential steps from trigger to final output?
3. What inputs does each step require?
4. What decisions are made at each step? (if/then logic)
5. Where are the bottlenecks or error-prone points?
6. What's the final deliverable and who receives it?

Present this as a numbered workflow with decision points marked.
</blockquote>

'''Phase 2 — Design the AI workflow.''' Send:

<blockquote>
Now redesign this as an AI-automated workflow. For each step, specify:

| Step || Human or AI? || If AI: what prompt template? || If Human: what decision? || Input || Output || Quality gate

Rules:
- Some steps should remain human (judgment calls, approvals, sensitive decisions)
- Every AI step needs a quality gate — how do you know the output is good enough to proceed?
- Include error handling — what happens when an AI step produces bad output?
- Include a feedback mechanism — how does the workflow improve over time?
</blockquote>

'''Phase 3 — Write the prompt templates.''' For each AI step in the workflow:

<blockquote>
Write the production-ready prompt template for Step '''[N]''': '''[step name]'''

The template should include:
- Role definition for the AI
- Clear input specification with [PLACEHOLDERS]
- Exact output format requirements
- Quality criteria the output must meet
- An example of good output vs. bad output

This prompt should work reliably every time with different inputs. It should be usable by someone who didn't design the workflow.
</blockquote>

'''Phase 4 — Run the workflow end-to-end.''' Execute the full pipeline with real data. Track:
* Time per step (manual vs. AI-assisted)
* Quality gate pass/fail rates
* Where you had to intervene or override
* Total time saved vs. the manual process

'''Phase 5 — Document the blueprint.''' Create a 1-page workflow document:

<blockquote>
Write a "Workflow Blueprint" for this process that includes:
1. '''Trigger:''' What starts the workflow
2. '''Flow diagram:''' Step-by-step with decision points (use text-based flowchart)
3. '''Prompt templates:''' Reference to each template (step number and name)
4. '''Quality gates:''' What to check at each stage
5. '''Error handling:''' What to do when something fails
6. '''Maintenance:''' How to update the workflow as requirements change
7. '''Metrics:''' How to measure whether the workflow is working well
</blockquote>

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a process''' — Pick something that takes 30+ minutes, involves multiple steps, and happens regularly. The more manual the current process, the bigger the payoff.
# '''Map the current workflow''' — Document every step, decision point, and handoff. You can't automate what you don't understand.
# '''Design the hybrid workflow''' — Decide what AI handles vs. what stays human. Add quality gates and error handling. Not everything should be automated.
# '''Build the prompt templates''' — Write production-grade prompts for each AI step. These should be reusable by anyone, not just you.
# '''Test end-to-end''' — Run the full workflow with real data. Measure time, quality, and failure points.
# '''Document the blueprint''' — Create a shareable document that anyone could use to run this workflow.

'''"Done" looks like:''' A complete, tested workflow blueprint with prompt templates, quality gates, and measured time savings. Something you could hand to a colleague and they could execute without additional explanation.

----

== 🧭 Why this matters (Strategists start here) ==

In [[The Prompt Chain|WA-Intermediate-01]], you built a 3-step prompt chain. Here, you're building a '''production-grade workflow''' — the kind of thing that saves hours per week and can be delegated. The key differences from an intermediate prompt chain: quality gates (not just chaining outputs blindly), error handling (what happens when AI fails), and documentation (others can run it without you). This is directly transferable to tools like n8n, Make, or Zapier with AI steps. The blueprint format is also the deliverable that organizations pay consultants to produce.

----

== Reflection ==
* How much time did the automated workflow save compared to the manual process? Is the saving worth the design effort?
* Which quality gates caught real problems? Which were unnecessary overhead?
* Where did AI fail and require human override? Was that predictable from the design phase, or did it only emerge during testing?
* 💬 ''Walk a colleague through your workflow blueprint and ask them to find the step most likely to fail. Fresh eyes spot single points of failure you've normalized.'' (Social Learners)

== ⬆️ Level up ==

You've reached the advanced level for Workflow Automation. From here, consider:
* Implementing this workflow in an automation tool (n8n, Make, Zapier) for true hands-free execution
* Combining this with [[Design Your Agent Workflow|AC-Advanced-01]] to add multi-agent architecture to your workflow steps
* Measuring workflow performance over 4 weeks and iterating based on failure data

Back to [[Workflow Automation|Workflow Automation]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Workflow Automation Exercises]]

The Verification Checklist

2026-03-16T16:28:22Z

Admin: Fix 2 internal link(s)

''Build a personal AI verification checklist and stress-test it against real AI outputs. 25 minutes.''

<blockquote>
'''One-liner:''' Build a personal AI verification system — a checklist you'll actually use — and stress-test it against real AI outputs to find its limits.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

You're going to build a verification checklist, then immediately try to break it.

'''Step 1 — Generate something to verify.''' Ask AI to produce a piece of content you might actually use in your work:

<blockquote>
Write a '''[deliverable type — e.g., client email, project proposal, market analysis, technical recommendation]''' about '''[topic relevant to your work]'''. Make it detailed and specific. Include data points, recommendations, and reasoning.
</blockquote>

'''Step 2 — Build your checklist.''' Before reading the output carefully, write your own verification checklist. Start with these categories and add your own:

| Check || Question || Pass/Fail

| '''Factual claims''' || Are specific numbers, dates, or statistics verifiable? ||
| '''Sources''' || Could I find the original source for any cited information? ||
| '''Reasoning''' || Does the logic hold? Are there hidden assumptions? ||
| '''Completeness''' || What important perspective or consideration is missing? ||
| '''Tone/audience''' || Is the tone appropriate? Would the intended audience trust this? ||
| '''Actionability''' || Are the recommendations specific enough to actually follow? ||
| '''Your domain check''' || [Add a check specific to your field] ||
| '''Your domain check''' || [Add another check specific to your field] ||

'''Step 3 — Apply the checklist.''' Go through the AI output line by line using your checklist. Mark each check as pass or fail. For every fail, note what the issue is.

'''Step 4 — Stress-test the checklist.''' Now deliberately ask AI to produce something harder to verify:

<blockquote>
Write the same type of '''[deliverable]''' but on a topic I'm less familiar with: '''[topic outside your expertise]'''. Make it equally detailed and authoritative.
</blockquote>

Apply your checklist again. Where does it fail to catch problems? What check do you need to add?

'''Step 5 — Finalize.''' Update your checklist based on what you learned. Save it where you'll actually use it — bookmark it, pin it, print it, whatever works.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Generate test content''' — Ask AI to produce a work-relevant deliverable. This gives you realistic material to verify.
# '''Draft your checklist''' — Build a structured verification list covering factual accuracy, reasoning quality, completeness, tone, and domain-specific concerns.
# '''Apply to familiar territory''' — Use the checklist on AI output about a topic you know. This lets you calibrate how well the checklist catches real errors.
# '''Apply to unfamiliar territory''' — Use the checklist on AI output about a topic you ''don't'' know well. This exposes gaps in your process — the errors you can only catch with domain knowledge.
# '''Iterate and save''' — Update the checklist based on what it missed. Save it in a format you'll actually reach for.

'''"Done" looks like:''' A tested, refined verification checklist (8-12 items) saved in a usable format, with evidence of at least one error it caught and one gap you identified and fixed.

----

== 🧭 Why this matters (Strategists start here) ==

In [[The Fact-Check Habit|EP-Basic-01]], you built a simple 3-question verification prompt. Here, you're building a '''systematic process''' — a checklist that works regardless of topic, catches both factual and reasoning errors, and is tuned to your specific work context. The community's 75% Ethical Prompting score means most people ''intend'' to verify AI output but lack a consistent method. A checklist turns good intentions into reliable behavior. At the advanced level, you'll scale this into a governance framework for a team; this exercise builds the individual practice first.

----

== Reflection ==
* Which check caught the most problems? Which was least useful?
* How did your verification experience change between the familiar topic and the unfamiliar one?
* Is your checklist something you'd actually pull up before sending an AI-generated deliverable? What format makes it most likely you'll use it?
* 💬 ''Trade checklists with a colleague. Have them apply yours to an AI output from their work — their feedback will reveal blind spots specific to your domain.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[The AI Governance Playbook|EP-Advanced-01]] — where you'll design an AI governance framework for a team or project.

Back to [[Pillars/Ethical Prompting|Ethical Prompting & Judgment]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Ethical Prompting Exercises]]

The Tinkerer

2026-03-16T16:28:21Z

Admin: Fix 4 internal link(s)

''The Tinkerer archetype — hands-on learners who learn best by doing, experimenting, and iterating.''

== How You Learn ==

You learn by doing. When you encounter a new AI tool or technique, your instinct is to open it up and start experimenting. You'd rather figure things out through trial and error than read a manual first. This makes you fast to adopt new tools and quick to discover what works — and what doesn't.

'''42% of AI Skills Quiz takers are Tinkerers''' — the most common learning style in the community.

== Your Strengths ==
* '''Fast experimentation.''' You try things while others are still reading about them. This gives you hands-on experience that no amount of theory can replace.
* '''Comfort with failure.''' You're not afraid of getting a bad AI output. You iterate, adjust, and try again — which is exactly how you get better at working with AI.
* '''Practical instinct.''' You naturally gravitate toward techniques that actually work in real situations, not just techniques that sound impressive.

== Where You Can Grow ==
* '''Pausing to reflect.''' Your speed is an asset, but sometimes the most valuable learning happens when you stop and ask "why did that work?" or "what pattern am I seeing?"
* '''Building repeatable processes.''' You might solve the same problem differently every time. The next level is turning your experiments into reusable templates and workflows.
* '''Sharing what you've learned.''' Your experimentation generates a lot of practical knowledge — but it stays in your head unless you document it.

== Recommended Exercises ==

Start with exercises that let you jump in immediately:
* [[The Reusable Prompt|The Reusable Prompt]] — Turn your tinkering into something repeatable
* [[The Stolen Technique|The Stolen Technique]] — Borrow a technique from another field (right up your alley)
* [[Your First AI Team Meeting|Your First AI Team Meeting]] — Experiment with multiple AI perspectives

== Your Entry Point ==

In every exercise, look for the '''"Jump in"''' section — it's designed for you. Start with the hands-on challenge, then circle back to the context and reflection.

== Recommended Pathway ==

If you're new to structured AI learning, try [[Pathway: Starting from Scratch|Starting from Scratch]] — it's designed to channel your experimental energy into lasting skills.

[[Category:AI Fluency Playbook]]
[[Category:Learner Archetypes]]

The Strategist

2026-03-16T16:28:20Z

Admin: Fix 4 internal link(s)

''The Strategist archetype — big-picture thinkers who need to understand the why before engaging with the how.''

== How You Learn ==

You want to understand the "why" before the "how." When you encounter a new AI tool or technique, your first question is "what's the big picture?" and "how does this fit into my work and career?" You think in terms of impact, value, and long-term direction.

'''23% of AI Skills Quiz takers are Strategists.'''

== Your Strengths ==
* '''Big-picture thinking.''' You see how AI fits into the broader context of your work, team, and industry. This means you make better decisions about which AI skills to invest in.
* '''Value-driven focus.''' You don't just learn for the sake of learning — you connect every skill to a real outcome. This makes your AI fluency practical and purposeful.
* '''Leadership instinct.''' You naturally think about how AI affects not just your work but your team's and organization's work. This positions you to guide others.

== Where You Can Grow ==
* '''Getting your hands dirty.''' Strategic thinking is essential, but AI fluency also requires practical experience. At some point, you need to move from "thinking about AI" to "working with AI."
* '''Starting small.''' Your instinct is to design the big system, but sometimes the most strategic thing you can do is master a single, small technique and build from there.
* '''Technical depth.''' You might be tempted to stay at the conceptual level. Pushing into the details of how AI actually works (prompt engineering, agent workflows) will make your strategic insights sharper.

== Recommended Exercises ==

Start with exercises that connect to bigger outcomes:
* [[The AI Governance Playbook|The AI Governance Playbook]] — Design team-level AI guidelines (strategic impact)
* [[The Signal in the Noise|The Signal in the Noise]] — Extract meaning from AI output (a skill that multiplies everything else)
* [[The Workflow Blueprint|The Workflow Blueprint]] — Design a complete AI-assisted workflow

== Your Entry Point ==

In every exercise, look for the '''"Why this matters"''' section — it's designed for you. Understand the strategic context and career value before diving into the steps.

== Recommended Pathway ==

If you want a structured route, try [[Pathway: High Synthesis, Low Agent Collaboration|High Synthesis, Low Agent Collaboration]] — it builds from your analytical strength into more hands-on agent work.

[[Category:AI Fluency Playbook]]
[[Category:Learner Archetypes]]

The Stolen Technique

2026-03-16T16:28:19Z

Admin: Fix 1 internal link(s)

''Take an AI technique from a completely different field and apply it to your own work. 15 minutes.''

<blockquote>
'''One-liner:''' Take an AI technique from a completely different field and apply it to your own work — discovering that the best prompting ideas are often borrowed.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a field that is '''not''' your own. If you work in marketing, pick engineering. If you're a designer, pick finance. If you're a developer, pick journalism. The more unfamiliar, the better.

'''Step 1 — Discover a technique.''' Send this prompt:

<blockquote>
How do professionals in '''[unfamiliar field]''' use AI in their daily work? Give me 5 specific, concrete techniques — not general concepts. For each technique, describe: what they prompt the AI to do, what input they provide, and what output they get. Focus on techniques that are unique to this field.
</blockquote>

'''Step 2 — Steal the best one.''' Pick the technique that seems most interesting or most different from how you currently use AI. Then send:

<blockquote>
I work in '''[your field]'''. Take the technique you described as #[number] — '''[briefly describe it]''' — and help me adapt it for my work. Specifically:
1. What would the equivalent input look like in my field?
2. How would I modify the prompt to fit my context?
3. What output would I expect?
4. Write me a ready-to-use prompt that applies this borrowed technique to '''[a specific task you do]'''.
</blockquote>

'''Step 3 — Test it.''' Copy the adapted prompt. Use it on a real task. Compare the result to how you'd normally approach it.

'''Example — a marketer borrowing from investigative journalism:'''

The technique: Journalists use AI to cross-reference claims across multiple sources and flag inconsistencies.

The adaptation: A marketer uses the same technique to cross-reference their product claims against competitor claims and customer reviews, flagging gaps between promise and reality.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Pick an unfamiliar field''' — Choose something genuinely outside your expertise. The discomfort is the point — that's where non-obvious ideas live.
# '''Research AI techniques in that field''' — Use AI to discover how professionals in that domain use AI tools. Look for specific techniques, not generalities.
# '''Identify a transferable technique''' — Pick one that solves a problem similar to something in your work, even though it looks completely different on the surface.
# '''Adapt with AI's help''' — Ask the AI to bridge the gap between the source domain and your domain. Get a ready-to-use prompt.
# '''Test the borrowed technique''' — Apply it to a real task and evaluate whether it gives you a different (and possibly better) result than your usual approach.

'''"Done" looks like:''' You have a working prompt borrowed from another field that gives you a new angle on a familiar task.

----

== 🧭 Why this matters (Strategists start here) ==

Most people prompt AI using patterns from their own field — but the most powerful AI techniques are often domain-agnostic. Researchers structure AI analysis differently than marketers, engineers test AI outputs differently than writers, and each field has developed prompting patterns the others rarely see. Cross-domain reframing is how you break out of local optima in your AI usage. At the intermediate level, you'll systematically adapt entire prompt strategies across domains; this exercise builds the muscle of looking outside your field for AI inspiration.

----

== Reflection ==
* Did the borrowed technique produce a noticeably different result than your usual approach? Better, worse, or just different?
* What made the technique transferable? Was it the structure, the question type, or the underlying problem it solves?
* Which other field would you explore next for AI techniques? What made you choose it?
* 💬 ''Ask a colleague from a different department how they use AI. You'll likely discover a technique you've never considered — that's cross-domain reframing in action.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[The Framework Transplant|CDR-Intermediate-01]] — where you'll systematically adapt an entire prompting strategy from an unfamiliar domain.

Back to [[Cross-Domain Reframing|Cross-Domain Reframing]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Cross-Domain Reframing Exercises]]

The Social Learner

2026-03-16T16:28:18Z

Admin: Fix 4 internal link(s)

''The Social Learner archetype — collaborative learners who build understanding through discussion and shared experience.''

== How You Learn ==

You learn best through conversation and collaboration. When you encounter a new AI tool or technique, your instinct is to discuss it with someone — a colleague, a community, or even the AI itself. Ideas crystallize for you through dialogue, not isolation.

'''6% of AI Skills Quiz takers are Social Learners''' — the rarest learning style, but one with unique strengths.

== Your Strengths ==
* '''Collaborative instinct.''' You naturally think about AI in terms of how it affects teams, communication, and shared work. This gives you an edge in agent collaboration and multi-perspective exercises.
* '''Teaching ability.''' Explaining things to others is how you deepen your own understanding. This makes you a natural AI fluency advocate in your team.
* '''Critical dialogue.''' You're good at questioning AI output in conversation — bouncing ideas off others, challenging assumptions, and building shared understanding. This is the essence of ethical prompting.

== Where You Can Grow ==
* '''Independent practice.''' Some AI skills need to be built through solo practice — prompt engineering, workflow design, synthesis. Finding ways to build these skills even when you don't have a discussion partner will accelerate your growth.
* '''Moving from discussion to action.''' Your conversations about AI generate great ideas, but the next level is turning those ideas into concrete exercises, templates, and workflows.
* '''Deep technical skills.''' You might rely on others' technical expertise in group settings. Building your own hands-on comfort with AI tools will make your collaborative contributions even more valuable.

== Recommended Exercises ==

Start with exercises that involve multiple perspectives:
* [[Your First AI Team Meeting|Your First AI Team Meeting]] — Work with AI in multiple roles (this is your exercise)
* [[The Multi-Source Brief|The Multi-Source Brief]] — Triangulate multiple AI perspectives
* [[The Fact-Check Habit|The Fact-Check Habit]] — Build verification skills you can discuss with colleagues

== Your Entry Point ==

In every exercise, look for the '''"Discuss"''' section in the Reflection — it's designed for you. Use the prompts to start a conversation with a colleague or community member.

== Recommended Pathway ==

If you're looking for a guided route, try [[Pathway: Starting from Scratch|Starting from Scratch]] — it gives you structured exercises you can do solo, with plenty of reflection prompts to discuss with others.

[[Category:AI Fluency Playbook]]
[[Category:Learner Archetypes]]

The Signal in the Noise

2026-03-16T16:28:18Z

Admin: Fix 1 internal link(s)

''Turn a messy AI brainstorm into structured, actionable insight. 15 minutes.''

<blockquote>
'''One-liner:''' Turn a messy AI brainstorm into a structured, actionable insight — learning to extract what matters and discard what doesn't.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a topic you're genuinely curious about or working on. It could be a business challenge, a learning goal, or a decision you need to make.

'''Step 1 — Generate the mess.''' Send this prompt to any AI:

<blockquote>
Brainstorm 15-20 ideas about '''[your topic]'''. Don't filter or organize — just generate as many ideas as possible, even contradictory or half-formed ones. Number each idea.
</blockquote>

'''Step 2 — Extract the signal.''' Now send this follow-up:

<blockquote>
Look at the brainstorm you just generated. Identify:
1. '''The top 3 ideas''' that are most actionable within the next week
2. '''The 1 idea''' that's most surprising or non-obvious
3. '''The 2 ideas''' that contradict each other — and what the tension between them reveals
4. '''The pattern''' — what theme or assumption connects most of these ideas?

For each, explain your reasoning in one sentence.
</blockquote>

'''Step 3 — Challenge the synthesis.''' Send this:

<blockquote>
Now tell me what's missing from this brainstorm. What obvious angle or perspective did you fail to include? Add 3 ideas that fill that gap.
</blockquote>

Read the final output. You started with noise; you now have structured insight. The skill here isn't prompting — it's knowing what questions to ask ''after'' the AI generates raw material.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a topic''' — Something you care about. The exercise works best with real problems, not hypotheticals.
# '''Generate raw material''' — Ask AI for a large, unfiltered brainstorm (15-20 ideas). The messier the better — that's the point.
# '''Apply a synthesis framework''' — Use the structured follow-up prompt to force the AI to categorize, rank, and find patterns in its own output.
# '''Identify gaps''' — Ask the AI what it missed, then evaluate whether the gap-filling ideas actually change your understanding.
# '''Capture your insight''' — Write a single sentence summarizing what you learned that you didn't know before.

'''"Done" looks like:''' You have 3 actionable ideas, 1 non-obvious insight, a clear tension to think about, and a unifying pattern — extracted from a wall of brainstorm text.

----

== 🧭 Why this matters (Strategists start here) ==

AI is excellent at generating volume but mediocre at distinguishing signal from noise — that's still a human skill. This exercise builds your ability to use AI as a '''thinking amplifier''' rather than an answer machine. The synthesis framework (rank, surprise, contradict, pattern) is reusable: apply it to research outputs, meeting notes, customer feedback analysis, or any situation where you need to extract meaning from quantity. At the intermediate level, you'll synthesize across ''multiple'' AI outputs; this exercise builds the foundation.

----

== Reflection ==
* Did the AI's ranking match your instinct? Where did you disagree, and what does that tell you about the AI's priorities vs. yours?
* Was the "gap" the AI identified actually a meaningful blind spot, or was it filler?
* Would you use this brainstorm-then-synthesize pattern again? For what kinds of problems does it work best?
* 💬 ''Run the same brainstorm prompt with a colleague present. Compare which ideas you each gravitate toward — the difference reveals your respective assumptions.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[The Multi-Source Brief|IS-Intermediate-01]] — where you'll synthesize across multiple AI sessions to build a more complete picture.

Back to [[Insight Synthesis|Insight Synthesis]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Insight Synthesis Exercises]]

The Reusable Prompt

2026-03-16T16:28:17Z

Admin: Fix 1 internal link(s)

''Turn a repeatable task into a reusable AI prompt template. Your first step toward automation. 15 minutes.''

<blockquote>
'''One-liner:''' Turn a task you do repeatedly into a reusable AI prompt template that works every time — your first step toward automation.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Think of something you do at least once a week that involves writing, analyzing, or summarizing. Examples: writing a status update, summarizing meeting notes, drafting an email to a client, reviewing a document.

'''Step 1 — Do it once with AI.''' Open any AI chat and do the task the way you normally would — just ask the AI to help. Don't overthink the prompt. Just get the job done.

'''Step 2 — Reverse-engineer your prompt.''' Now send this:

<blockquote>
Look at the prompt I just gave you and the output you produced. Help me turn this into a '''reusable template''' that I can use every time I need to do this task. The template should have:
1. '''Clear placeholders''' — marked with [BRACKETS] for the parts that change each time
2. '''Fixed instructions''' — the parts that stay the same every time
3. '''Output format specification''' — exactly what the result should look like (length, structure, tone)

Write the template so someone else on my team could use it without any additional explanation.
</blockquote>

'''Step 3 — Test it.''' Copy the template. Start a new chat. Paste the template and fill in the placeholders with a different example of the same task. Does the output match the quality of your original? If not, adjust the template.

'''Here's a concrete example:'''

''Original task:'' "Summarize this meeting for my team"

''Reusable template:''
<blockquote>
Summarize the following meeting notes for a team update.

'''Meeting notes:''' [PASTE NOTES HERE]

'''Output requirements:'''
- Start with a 1-sentence summary of the main decision or outcome
- List action items with owner names in bold
- Flag any unresolved questions
- Keep the total summary under 150 words
- Tone: professional but informal
</blockquote>

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Identify a repeatable task''' — Pick something you do weekly that involves text: writing, summarizing, analyzing, or formatting. The more repetitive, the better.
# '''Do it once with AI''' — Complete the task normally. Don't try to be clever — just get a result you're happy with.
# '''Extract the template''' — Ask the AI to help you identify what's fixed (instructions, format, tone) vs. what changes (the input data). Build a reusable template with clear placeholders.
# '''Test with a new example''' — Use the template on a fresh instance of the same task. Compare quality to the original.
# '''Refine if needed''' — If the template didn't produce equally good output, identify what was missing and add it.

'''"Done" looks like:''' You have a saved prompt template with clear placeholders that consistently produces good output for your repeatable task.

----

== 🧭 Why this matters (Strategists start here) ==

Most people use AI in one-off conversations that disappear. This exercise introduces the shift from '''ad-hoc prompting to systematic workflows''' — the foundation of all AI automation. A reusable template is the simplest form of an AI workflow: defined input, consistent process, predictable output. At the intermediate level, you'll chain multiple templates together into multi-step workflows. Every automated AI process in production started as someone's reusable prompt.

----

== Reflection ==
* What did you have to add to the template that wasn't obvious from the original prompt?
* Did the template produce consistent quality with different inputs, or did you need to tweak it? What was missing?
* How much time will this template save you per week? Is it enough to justify the setup effort?
* 💬 ''Send your template to a colleague who does the same task. Can they use it without any explanation? Their confusion points reveal where the template needs more specificity.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[The Prompt Chain|WA-Intermediate-01]] — where you'll chain multiple prompt templates into a multi-step workflow.

Back to [[Workflow Automation|Workflow Automation]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Workflow Automation Exercises]]

The Research Pipeline

2026-03-16T16:28:16Z

Admin: Fix 3 internal link(s)

''Build a complete research synthesis pipeline with evidence grading and contradiction analysis. 40 minutes.''

<blockquote>
'''One-liner:''' Build a complete research synthesis pipeline — from question to evidence-graded conclusions — using structured AI queries and your own critical judgment.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a question you genuinely need answered for your work. Not a trivia question — something where the answer shapes a real decision.

'''Phase 1 — Define the research question.''' Send:

<blockquote>
I need to research this question: '''[your question]'''

Help me refine it into a research-ready question by:
1. Breaking it into 3-4 sub-questions that, if answered, would fully address the main question
2. For each sub-question, identifying what type of evidence would count as a strong answer (data, expert consensus, case studies, logical argument, etc.)
3. Flagging any assumptions embedded in the main question that I should test
</blockquote>

'''Phase 2 — Structured evidence gathering.''' For each sub-question, run a separate AI query:

<blockquote>
Research sub-question: '''[sub-question]'''

For this query, I want structured evidence:
- '''Strong evidence:''' Claims supported by widely documented data, peer-reviewed research, or established expert consensus
- '''Moderate evidence:''' Claims supported by credible case studies, industry reports, or respected analysis
- '''Weak evidence:''' Claims based on anecdotes, single examples, logical inference without data, or common assertions that may not hold up

Classify every claim you make. If you're not sure about the evidence quality, say so. I'd rather have honest uncertainty than false confidence.
</blockquote>

'''Phase 3 — Contradiction analysis.''' After running all sub-queries, send this to a fresh session:

<blockquote>
Here are the findings from my research on '''[main question]''':

'''Sub-question 1 findings:''' [paste summary]
'''Sub-question 2 findings:''' [paste summary]
'''Sub-question 3 findings:''' [paste summary]

Analyze the contradictions:
1. Where do the findings from different sub-questions conflict?
2. Which conflicts can be resolved by looking at the evidence quality?
3. Which conflicts are genuine unresolved tensions?
4. What additional evidence would resolve the remaining tensions?
</blockquote>

'''Phase 4 — Your synthesis.''' Write a 500-word research brief yourself (not AI-generated) that answers your original question. Structure it as:
# '''Bottom line:''' Your answer in 1-2 sentences
# '''Key evidence:''' The 3 strongest pieces of evidence supporting your answer, with evidence grades
# '''Key uncertainty:''' What you're least confident about and why
# '''What would change your mind:''' 1-2 pieces of evidence that, if found, would reverse your conclusion

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Formulate a research question''' — Choose something decision-relevant. Use AI to decompose it into sub-questions with defined evidence standards.
# '''Gather evidence by sub-question''' — Run separate queries for each sub-question, requiring the AI to grade its own evidence quality (strong/moderate/weak).
# '''Analyze contradictions''' — Feed all findings into a fresh session and ask for conflict analysis. Identify which conflicts are real vs. caused by weak evidence.
# '''Write your own synthesis''' — Produce a 500-word brief that answers the question, cites evidence with quality grades, and states what would change your mind.
# '''Assess the pipeline''' — Evaluate whether this process produced a meaningfully better answer than a single AI query would have.

'''"Done" looks like:''' A research brief that clearly distinguishes strong from weak evidence, acknowledges uncertainty, and provides a decision-ready answer with stated confidence.

----

== 🧭 Why this matters (Strategists start here) ==

This exercise combines the skills from [[The Signal in the Noise|IS-Basic-01]] (extracting signal from noise) and [[The Multi-Source Brief|IS-Intermediate-01]] (triangulating across perspectives) into a '''complete research methodology'''. The evidence grading system prevents the common failure mode of treating all AI output as equally reliable. The contradiction analysis surfaces genuinely open questions rather than papering over them. This pipeline is directly applicable to due diligence, competitive intelligence, policy analysis, and any context where the cost of being wrong is high and the question is too complex for a single query.

----

== Reflection ==
* Did the evidence grading change which findings you trusted? Were you surprised by what was classified as "weak"?
* How did the contradiction analysis change your initial view?
* Was the 500-word synthesis harder or easier than expected? What was the hardest part — compression, confidence, or acknowledging uncertainty?
* 💬 ''Teach this pipeline to a colleague and have them run it on a different question. Compare how you each handle the "what would change your mind" step — that reveals different attitudes toward uncertainty.'' (Social Learners)

== ⬆️ Level up ==

You've reached the advanced level for Insight Synthesis. From here, consider:
* Using this pipeline for a real decision and tracking whether your evidence-graded conclusion held up
* Combining this with [[Design Your Agent Workflow|AC-Advanced-01]] to delegate different research phases to different agent roles
* Teaching this method to a colleague and seeing how they adapt it

Back to [[Insight Synthesis|Insight Synthesis]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Insight Synthesis Exercises]]

The Prompt Chain

2026-03-16T16:28:15Z

Admin: Fix 2 internal link(s)

''Build a multi-step AI workflow where each step's output feeds into the next. 25 minutes.''

<blockquote>
'''One-liner:''' Build a multi-step AI workflow where each step's output feeds into the next — turning a complex task into a repeatable pipeline.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a task that has at least 3 distinct phases. Examples: writing a blog post (research, outline, draft, edit), analyzing a dataset (clean, analyze, summarize, recommend), or preparing a presentation (topic research, slide structure, talking points, Q&A prep).

'''Build a 3-step chain.''' Each step is a separate prompt. The output of each step becomes the input of the next.

'''Step 1 — Research/Gather:'''
<blockquote>
You are a research assistant. Your job is to gather the raw material for '''[your task]'''.

Topic/context: '''[describe what you're working on]'''

Produce a structured collection of: key facts, relevant examples, important considerations, and any constraints. Organize by theme. Do not draft anything — just collect the ingredients.
</blockquote>

Copy the output. Start a new prompt (or clearly reset context).

'''Step 2 — Structure/Draft:'''
<blockquote>
You are a content architect. Your job is to turn raw research into a structured draft.

Here is the research material: '''[paste Step 1 output]'''

The final deliverable is: '''[describe what you need — a blog post, a report, a strategy doc, etc.]'''

Create a structured draft. Include clear sections, key arguments in order, and placeholders for any examples or data points from the research. Focus on logical flow and completeness.
</blockquote>

Copy the output. Start a new prompt.

'''Step 3 — Polish/Critique:'''
<blockquote>
You are a senior editor. Your job is to make this draft publication-ready.

Here is the draft: '''[paste Step 2 output]'''

The audience is: '''[describe who will read this]'''

Do three things:
1. Improve clarity — simplify any convoluted sentences, cut unnecessary words
2. Strengthen weak points — flag any claim that needs better support and add it
3. Check consistency — ensure tone, terminology, and formatting are uniform throughout

Produce the final version with an editor's note listing your key changes.
</blockquote>

'''Now document the chain.''' Write down the 3 prompts as a reusable template (with [PLACEHOLDERS] for the parts that change). You've just built a prompt pipeline.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a multi-phase task''' — Something that naturally has distinct stages (research → create → refine). The more phases, the more the chain helps.
# '''Design the chain''' — Write 3 prompts, each with a clear role, input expectation, and output format. The key constraint: each step's output must contain everything the next step needs.
# '''Run the chain''' — Execute each step sequentially, passing the output forward. Use fresh contexts between steps to prevent bleed-through.
# '''Evaluate information flow''' — Notice where context was lost between steps. What did Step 3 need that Step 2 didn't preserve?
# '''Document as a template''' — Save the chain with placeholders so you can reuse it for the same type of task.

'''"Done" looks like:''' A completed deliverable that went through a 3-step pipeline, plus a documented prompt chain template with placeholders for reuse.

----

== 🧭 Why this matters (Strategists start here) ==

In [[The Reusable Prompt|WA-Basic-01]], you built a single reusable prompt. Here, you're learning to '''chain prompts into a workflow''' — the building block of all production AI automation. Every AI-powered pipeline (content generation, data analysis, document processing) is fundamentally a prompt chain with handoffs. The skill you're building — decomposing a task into stages, defining clear inputs and outputs, managing context between steps — is the same skill used in tools like n8n, Zapier AI, or custom LLM pipelines. Manual chaining teaches you what to automate and where the bottlenecks live.

----

== Reflection ==
* Where did context get lost between steps? What information did a later step need that an earlier step didn't pass along?
* Did the 3-step chain produce better output than a single "do everything" prompt? Where specifically was the improvement?
* Which step in the chain was the weakest link? How would you redesign it?
* 💬 ''Have a colleague run your documented chain on a different task of the same type. Their experience reveals whether your chain is truly reusable or depends on your implicit knowledge.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[The Workflow Blueprint|WA-Advanced-01]] — where you'll design and document a complete AI-automated workflow for a business process.

Back to [[Workflow Automation|Workflow Automation]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Workflow Automation Exercises]]

The Planner

2026-03-16T16:28:14Z

Admin: Fix 4 internal link(s)

''The Planner archetype — structured, methodical learners who prefer to understand the plan before executing.''

== How You Learn ==

You like to understand the landscape before you act. When you encounter a new AI tool or technique, your first instinct is to read the instructions, understand the structure, and map out your approach. You prefer clear steps and predictable outcomes.

'''25% of AI Skills Quiz takers are Planners.'''

== Your Strengths ==
* '''Systematic approach.''' You build processes that work reliably, not just once but every time. This makes you excellent at workflow automation and creating templates others can follow.
* '''Thoroughness.''' You catch edge cases and think about what could go wrong before it does. This makes your AI-assisted work more dependable.
* '''Documentation instinct.''' You naturally organize what you learn, which means your insights don't get lost — and they're easy to share.

== Where You Can Grow ==
* '''Getting started sooner.''' Your desire to plan can sometimes delay action. With AI, the feedback loop is so fast that trying something "imperfect" often teaches you more than planning the perfect approach.
* '''Embracing ambiguity.''' AI doesn't always give predictable results. Learning to work with uncertainty — and even leverage it — is a key growth area.
* '''Cross-domain exploration.''' Your structured thinking keeps you effective, but sometimes the most valuable AI insights come from unexpected places. Try borrowing techniques from unfamiliar fields.

== Recommended Exercises ==

Start with exercises that reward structured thinking:
* [[The Fact-Check Habit|The Fact-Check Habit]] — Build a verification process (you'll love the systematic approach)
* [[The Reusable Prompt|The Reusable Prompt]] — Create a well-structured prompt template
* [[The Multi-Source Brief|The Multi-Source Brief]] — Triangulate AI perspectives into a clear brief

== Your Entry Point ==

In every exercise, look for the '''"Plan first"''' section — it's designed for you. Read the overview and structured preview before starting the hands-on work.

== Recommended Pathway ==

If you're ready to push into less familiar territory, try [[Pathway: Strong Communicator, Building Technical Confidence|Strong Communicator, Building Technical Confidence]] — it bridges your organizational strengths into more technical AI skills.

[[Category:AI Fluency Playbook]]
[[Category:Learner Archetypes]]

The Multi-Source Brief

2026-03-16T16:28:13Z

Admin: Fix 2 internal link(s)

''Synthesize outputs from three separate AI perspectives into a single coherent analysis. 25 minutes.''

<blockquote>
'''One-liner:''' Synthesize outputs from three separate AI queries into a single coherent analysis — building the skill of triangulating AI perspectives.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a question or topic you need to actually understand — a market trend, a technology choice, a strategic decision, a complex issue in your field.

'''Run three separate queries''' in three different AI sessions (or clear context between each). Each query approaches the same topic from a different angle:

'''Query 1 — The Optimist:'''
<blockquote>
Analyze '''[your topic]''' from the most optimistic perspective. What's the strongest case that this will succeed/matter/grow? Cite specific evidence, trends, and examples. Be persuasive, not balanced.
</blockquote>

'''Query 2 — The Skeptic:'''
<blockquote>
Analyze '''[your topic]''' from a skeptical perspective. What's the strongest case that this is overhyped, risky, or likely to fail? Cite specific evidence, counterexamples, and historical parallels where similar things didn't pan out. Be rigorous, not cynical.
</blockquote>

'''Query 3 — The Analyst:'''
<blockquote>
Analyze '''[your topic]''' by identifying the 3-5 key variables that will determine the outcome. Don't argue for or against — map the decision space. For each variable, describe what would need to be true for a positive outcome vs. a negative one. Include what we don't yet know.
</blockquote>

'''Now synthesize.''' Open a fresh document (not an AI chat). Write a 250-word brief that answers:
# '''What do all three perspectives agree on?''' (This is likely true.)
# '''Where do the Optimist and Skeptic directly contradict each other?''' (This is where the real uncertainty lives.)
# '''Which of the Analyst's key variables would resolve the contradiction?''' (This is what you need to investigate.)
# '''Your take''' — Given all three inputs, what's your position and what would change your mind?

The brief should be something you'd share with a colleague or decision-maker. No AI jargon, no meta-commentary about the process.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a topic''' — Pick something with genuine uncertainty. If the answer is obvious, the exercise won't stretch you. Good candidates: emerging trends, strategic choices, technology bets, or contested ideas in your field.
# '''Run three separate AI sessions''' — Optimist, Skeptic, and Analyst. Use fresh contexts (new chats or cleared conversations) so each query isn't influenced by the others.
# '''Read all three outputs''' — Don't start synthesizing until you've read all three. Notice your own bias — which perspective did you instinctively agree with?
# '''Write the synthesis yourself''' — This is the critical step. Don't ask AI to synthesize for you. The skill you're building is ''your'' ability to integrate contradictory information.
# '''Distill to 250 words''' — Force compression. A good brief is one where every sentence earns its place.

'''"Done" looks like:''' A 250-word brief you'd be comfortable sharing with a colleague, built from three distinct AI perspectives, with a clear statement of what you believe and what would change your mind.

----

== 🧭 Why this matters (Strategists start here) ==

In [[The Signal in the Noise|IS-Basic-01]], you extracted insights from a single AI output. Here, you're building a fundamentally harder skill: '''triangulating across multiple AI perspectives to form your own judgment'''. This is exactly what senior decision-makers do with human advisors — they don't take any single perspective at face value. The discipline of writing the synthesis yourself (rather than asking AI to do it) ensures you're developing the judgment, not outsourcing it. This skill directly applies to research, due diligence, competitive analysis, and any situation where multiple data sources tell different stories.

----

== Reflection ==
* Which perspective (Optimist, Skeptic, Analyst) was most useful? Which felt like filler?
* Did writing the synthesis yourself change your view compared to where you started? At what point in the writing did it shift?
* Would you share this brief with a decision-maker? If not, what's missing?
* 💬 ''Share your 250-word brief with someone who knows the topic. Ask them what they'd challenge — their pushback will tell you where your synthesis was weakest.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[The Research Pipeline|IS-Advanced-01]] — where you'll build a full research synthesis pipeline with structured evidence evaluation.

Back to [[Insight Synthesis|Insight Synthesis]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Insight Synthesis Exercises]]

The Handoff Protocol

2026-03-16T16:28:13Z

Admin: Fix 2 internal link(s)

''Split a problem across two separate AI sessions with different roles, then synthesize their outputs. 25 minutes.''

<blockquote>
'''One-liner:''' Split a problem across two separate AI sessions with different roles and contexts, then synthesize their outputs yourself — like managing a real team.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

You'll need two AI chat windows open at the same time (two browser tabs, or two different AI tools — either works).

'''Pick a project or decision''' that has at least two distinct dimensions. For example: "Create a content strategy for launching our new product."

'''Chat A — The Strategist.''' Open your first chat and send:

<blockquote>
You are a '''brand strategist''' with 15 years of experience. Your focus is positioning, audience targeting, and messaging clarity. You do NOT think about implementation details — that's someone else's job.

I'm working on: '''[your project]'''

Give me your strategic recommendations. Focus on: who the audience is, what the core message should be, and how to position this differently from competitors. Be specific and opinionated.
</blockquote>

'''Chat B — The Executor.''' Open your second chat and send:

<blockquote>
You are an '''operations-focused content producer'''. Your focus is practical execution: channels, formats, timelines, and resource requirements. You do NOT set strategy — you receive it and figure out how to make it real.

I'm working on: '''[your project]'''

Give me an execution plan. Focus on: which channels to prioritize, what content formats work best, a realistic timeline, and what resources I'll need. Be specific and practical.
</blockquote>

'''Now you're the manager.''' Read both outputs. Notice what Chat A assumed that Chat B would question, and vice versa. Then write your own synthesis:
* Where do these perspectives align?
* Where do they conflict?
* What did each one miss that the other caught?
* What's your actual plan, informed by both?

'''Optional bonus round:''' Take your synthesis and paste it back into one of the chats:

<blockquote>
Here's the combined strategy and execution plan I've built from two different advisors. Poke holes in it. What's still weak?
</blockquote>

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a project''' — Something real with both a strategic and practical dimension. Content launches, product decisions, event planning, and hiring processes all work well.
# '''Set up Chat A (Strategist)''' — Give it a clear strategic role with explicit boundaries. Tell it ''not'' to worry about implementation.
# '''Set up Chat B (Executor)''' — Give it a clear operational role with explicit boundaries. Tell it ''not'' to set strategy.
# '''Run both chats''' — Send the same project description to each, but with their respective role prompts.
# '''Synthesize manually''' — You are the integration point. Compare outputs, find gaps, resolve conflicts, and produce a combined plan.

'''"Done" looks like:''' You have a plan that neither AI session could have produced alone, and you can articulate what each perspective contributed.

----

== 🧭 Why this matters (Strategists start here) ==

In [[Your First AI Team Meeting|AC-Basic-01]], you simulated multiple perspectives in a single chat. Here, you're practicing a fundamentally different skill: '''managing separate agents with isolated contexts'''. This mirrors how real multi-agent systems work — each agent has a specific role, limited scope, and doesn't see the other's work. The human (you) acts as the orchestrator. This is the skill that scales: from two chats to entire AI-assisted workflows with specialized roles, handoff points, and quality gates.

----

== Reflection ==
* How did the outputs differ when each AI had a constrained role vs. a single AI doing both? Was the split worth the extra effort?
* What context got lost in the handoff between sessions? How would you design a better transfer summary?
* Did the synthesis step feel harder or easier than you expected? What made it difficult?
* 💬 ''Run this exercise with two colleagues, each managing one AI session. Compare the experience of synthesizing someone else's AI output vs. your own — it highlights how much implicit context lives in your head.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[Design Your Agent Workflow|AC-Advanced-01]] — where you'll design a complete multi-agent workflow with defined roles, handoffs, and feedback loops.

Back to [[Agent Collaboration|Agent Collaboration]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Agent Collaboration Exercises]]

The Framework Transplant

2026-03-16T16:28:12Z

Admin: Fix 2 internal link(s)

''Systematically transplant a problem-solving framework from another domain to solve your challenge. 25 minutes.''

<blockquote>
'''One-liner:''' Take a complete problem-solving framework from another domain and systematically adapt it to solve a challenge in your own work.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a challenge you're currently facing in your work — something you've been approaching the same way without breakthrough results.

'''Step 1 — Find a foreign framework.''' Send this prompt:

<blockquote>
I'm struggling with '''[your challenge]''' in my field of '''[your field]'''. I want a completely fresh approach. Give me 3 well-known problem-solving frameworks from '''different''' fields (engineering, medicine, military strategy, game design, ecology — anything outside my domain). For each framework:
1. Name and origin field
2. How it works (3-4 step process)
3. Why it might apply to my problem

Choose frameworks that are genuinely different from each other, not variations on the same idea.
</blockquote>

'''Step 2 — Deep-dive one framework.''' Pick the most promising or most surprising framework. Send:

<blockquote>
Let's go deeper on '''[chosen framework]'''. Walk me through how a professional in '''[its origin field]''' would apply this framework to a real problem in their domain. Be specific — give me a concrete example with actual steps, not abstractions.
</blockquote>

'''Step 3 — Systematic transplant.''' Now adapt it:

<blockquote>
Now help me transplant this framework to my challenge: '''[restate your challenge]'''.

Map each step of the framework to my context:
- '''Step 1 of framework''' → What does this look like in my situation?
- '''Step 2 of framework''' → What's the equivalent action?
- (continue for all steps)

For each mapping:
- What translates directly?
- What needs to be modified and how?
- What doesn't transfer at all, and what should replace it?

End with a concrete action plan I can execute this week.
</blockquote>

'''Step 4 — Stress test.''' Send:

<blockquote>
Play devil's advocate. Where does this transplanted framework break down when applied to my field? What assumptions from the original domain don't hold in mine? How should I adjust?
</blockquote>

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Identify your challenge''' — Pick something real where your current approaches have stalled. The exercise only works if you're genuinely stuck.
# '''Discover foreign frameworks''' — Use AI to surface structured problem-solving approaches from unfamiliar fields. Look for frameworks with clear steps, not just theories.
# '''Study the framework in its native context''' — Understand how it actually works in practice before trying to adapt it. This prevents shallow borrowing.
# '''Map step-by-step to your context''' — Systematically translate each step, noting where the mapping is direct, where it needs modification, and where it fails entirely.
# '''Stress test the adaptation''' — Identify where the transplant breaks down and adjust before committing to action.

'''"Done" looks like:''' A concrete action plan for your challenge, based on a framework from another field, with clear documentation of what translated, what was modified, and what was replaced.

----

== 🧭 Why this matters (Strategists start here) ==

In [[The Stolen Technique|CDR-Basic-01]], you borrowed a single technique from another field. Here, you're transplanting an '''entire framework''' — a much harder and more valuable skill. This is how breakthrough innovations happen: the structure of a solution transfers across domains even when the details don't. Toyota's production system was adapted from supermarket inventory management. Agile software development borrowed from lean manufacturing. The ability to systematically adapt frameworks across domains is what separates insight from coincidence. At the advanced level, you'll build an entire cross-domain prompt library; this exercise builds the adaptation methodology.

----

== Reflection ==
* Which parts of the framework transferred most easily? What does that tell you about the underlying structure of your problem?
* Where did the transplant break down? Was the breakdown due to domain differences, or did it reveal an assumption you hadn't questioned?
* Did the stress test change your action plan significantly, or just refine the edges?
* 💬 ''Explain the transplanted framework to someone in the original field. Their reaction ("that's not how we use it" or "interesting adaptation") tells you whether you captured the core principle or just the surface.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[The Cross-Domain Prompt Library|CDR-Advanced-01]] — where you'll build a cross-domain prompt library with documented transfer patterns.

Back to [[Cross-Domain Reframing|Cross-Domain Reframing]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Cross-Domain Reframing Exercises]]

The Fact-Check Habit

2026-03-16T16:28:11Z

Admin: Fix 1 internal link(s)

''Catch an AI making a confident mistake and build a simple verification process you'll use every time. 15 minutes.''

<blockquote>
'''One-liner:''' Catch an AI making a confident mistake — and build a simple verification process you'll use every time.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a topic you know well — your industry, your hobby, your area of expertise. Something where you can spot errors.

'''Step 1 — Get a confident answer.''' Send this prompt:

<blockquote>
Give me a detailed overview of '''[topic you know well]'''. Include specific facts, statistics, and examples. Be thorough and authoritative.
</blockquote>

Read the output carefully. '''Find at least one claim that feels off.''' It might be a statistic that seems too round, a date that feels wrong, a name that's slightly off, or a causal claim that oversimplifies reality.

'''Step 2 — Make the AI check itself.''' Send this:

<blockquote>
Look at your previous response. I want you to fact-check yourself. For each specific claim, statistic, or example you cited:
1. Rate your confidence (high / medium / low)
2. Flag anything you might have fabricated or estimated
3. Identify which claims are most likely to be wrong and why

Be ruthlessly honest. I'd rather know what you're uncertain about than have you defend everything.
</blockquote>

'''Step 3 — Verify.''' Pick the 2-3 claims the AI flagged as lowest confidence. Google them. Were they accurate, close but wrong, or completely fabricated?

'''Step 4 — Build your check.''' Based on what you just learned, write a 3-line "verification prompt" you can append to any AI output:

<blockquote>
Before I use this, tell me:
1. Which specific claims are you least confident about?
2. What did you estimate or approximate vs. know with certainty?
3. What should I verify independently before sharing this?
</blockquote>

Save this somewhere you'll see it. Use it as a default follow-up to any AI output you plan to rely on.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a familiar topic''' — You need to be able to spot errors, so pick something in your area of knowledge. Don't use an unfamiliar topic — you won't know what to verify.
# '''Generate an authoritative-sounding response''' — Ask AI for a detailed, factual overview. The more specific and confident the output, the more likely it contains subtle errors.
# '''Ask AI to fact-check itself''' — Use the self-audit prompt to force the AI to rate its own confidence and flag potential fabrications.
# '''Independently verify''' — Pick the lowest-confidence claims and check them against reliable sources. Track what was right, close, and wrong.
# '''Create your verification template''' — Build a reusable 3-question follow-up that you'll use after any AI output you plan to act on.

'''"Done" looks like:''' You've caught at least one AI error, you understand ''why'' the AI got it wrong, and you have a saved verification prompt you can use going forward.

----

== 🧭 Why this matters (Strategists start here) ==

The community's Ethical Prompting score is 75% — the highest of all five pillars. Most people ''know'' they should verify AI output, but few have a systematic process for doing so. This exercise closes the gap between awareness and practice by giving you a concrete, reusable tool. The verification prompt you build here becomes a habit — a 30-second step that catches errors before they become problems. At the intermediate level, you'll build a more comprehensive verification checklist; this exercise establishes the baseline behavior.

----

== Reflection ==
* What type of error did the AI make — a fabricated fact, a wrong date, or a subtle logical leap? Does the category matter for how you'd catch it?
* Did the AI's self-assessment match what you found when you verified manually? Was it too confident, too cautious, or well-calibrated?
* Will you actually use your verification prompt going forward? What would make it stick as a habit vs. something you forget about?
* 💬 ''Share a specific AI error you caught with a colleague. Ask them how they currently verify AI output — you may discover they don't.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[The Verification Checklist|EP-Intermediate-01]] — where you'll build a comprehensive verification checklist and stress-test it against real AI outputs.

Back to [[Pillars/Ethical Prompting|Ethical Prompting & Judgment]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Ethical Prompting Exercises]]

The Cross-Domain Prompt Library

2026-03-16T16:28:10Z

Admin: Fix 3 internal link(s)

''Build a documented library of prompt patterns borrowed from 3+ different fields. 40 minutes.''

<blockquote>
'''One-liner:''' Build a documented library of prompt patterns borrowed from 3+ different fields, with tested adaptations and transfer notes for your own domain.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

You're going to build a personal prompt library of techniques stolen from other fields — documented well enough to teach someone else.

'''Phase 1 — Survey 3 domains.''' Pick 3 fields that are different from your own ''and'' from each other. Send this to three separate sessions:

<blockquote>
How do professionals in '''[field]''' use AI in sophisticated ways? I don't want generic "they use ChatGPT" answers. Give me 5 advanced AI techniques or prompting patterns that are specific to this field. For each:
1. Name the technique
2. Describe the prompt pattern (what input, what instructions, what output format)
3. Why this technique works in this domain (what problem it solves)
4. Example prompt (ready to use)
</blockquote>

'''Phase 2 — Select your top 5.''' From the 15 techniques across 3 domains, pick the 5 that are most interesting or most likely to transfer. For each one, send:

<blockquote>
Analyze this technique from '''[source domain]''': '''[technique description]'''

Map the transfer potential:
1. '''Core principle:''' What's the underlying mechanism that makes this work, independent of domain?
2. '''Direct transfer:''' What would this look like applied to '''[your field]''' with minimal modification?
3. '''Modified transfer:''' What would need to change to make it work well in my context?
4. '''What doesn't transfer:''' What aspect is domain-specific and should be replaced?
5. '''Adapted prompt:''' Write a ready-to-use version for my field
</blockquote>

'''Phase 3 — Test each adapted prompt.''' Run all 5 adapted prompts on real tasks in your work. For each, document:

| Technique || Source Domain || My Task || Result Quality (1-5) || What Worked || What Needed Adjustment

'''Phase 4 — Build the library entry.''' For the 3 best-performing techniques, create a library card:

<blockquote>
Create a "Prompt Library Card" for this technique:

'''Name:''' [give it a memorable name]
'''Borrowed from:''' [source domain]
'''Core principle:''' [1 sentence — why this works]
'''Original use:''' [what it does in the source domain]
'''My adaptation:''' [what it does in my domain]
'''Ready-to-use prompt:'''
``<code>
[the tested, refined prompt with placeholders]
</code>``
'''When to use:''' [scenarios where this technique is the right choice]
'''When NOT to use:''' [scenarios where it fails or is overkill]
'''Transfer notes:''' [what I learned about adapting this — tips for others]
</blockquote>

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Survey 3 unfamiliar domains''' — Discover advanced AI techniques in three different fields. Cast a wide net — diversity of domains matters more than depth.
# '''Select the top 5 candidates''' — From 15 techniques, choose 5 based on transfer potential and novelty. Look for techniques that solve a problem ''structurally'' similar to one in your work.
# '''Analyze transfer mechanics''' — For each technique, separate the domain-specific elements from the core principle. Identify what transfers directly, what needs modification, and what should be replaced.
# '''Test all 5 adaptations''' — Run each adapted prompt on a real task. Document quality, surprises, and adjustments needed.
# '''Document the top 3''' — Create library cards with enough detail for someone else to use the technique without your guidance.

'''"Done" looks like:''' A 3-entry prompt library with tested techniques from other domains, complete with ready-to-use prompts, usage guidance, and transfer notes.

----

== 🧭 Why this matters (Strategists start here) ==

In [[The Stolen Technique|CDR-Basic-01]], you borrowed a single technique. In [[The Framework Transplant|CDR-Intermediate-01]], you transplanted an entire framework. Here, you're building a '''systematic practice''' — a personal library that compounds over time. The library card format forces you to articulate ''why'' a technique transfers, which is the meta-skill: once you can spot the structural similarity between domains, you can generate new cross-domain adaptations on your own. This library also becomes a shareable team asset — a collection of non-obvious AI techniques that others in your field won't have discovered.

----

== Reflection ==
* Which of the 3 domains produced the most transferable techniques? Why?
* Did any technique work ''better'' in your domain than in its original domain? What does that tell you?
* What pattern do you notice in what transfers well vs. what doesn't? Can you articulate a rule of thumb?
* 💬 ''Share your prompt library with colleagues and have them test the techniques on their own tasks. The techniques that work across multiple people's contexts are genuinely domain-agnostic — those are your keepers.'' (Social Learners)

== ⬆️ Level up ==

You've reached the advanced level for Cross-Domain Reframing. From here, consider:
* Expanding the library monthly — add one new cross-domain technique per month from a new field
* Sharing the library with colleagues and collecting their transfer notes
* Combining this with [[The Workflow Blueprint|WA-Advanced-01]] to build cross-domain techniques into automated workflows

Back to [[Cross-Domain Reframing|Cross-Domain Reframing]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Cross-Domain Reframing Exercises]]

The AI Governance Playbook

2026-03-16T16:28:09Z

Admin: Fix 2 internal link(s)

''Design a practical AI governance framework for a team or project. 40 minutes.''

<blockquote>
'''One-liner:''' Design a practical AI governance framework for a team or project — covering when to use AI, how to verify outputs, and what requires human judgment.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a real team, project, or organization you work with. You're going to design an AI usage framework they could actually adopt.

'''Step 1 — Map the AI touchpoints.''' Send this prompt:

<blockquote>
I'm designing an AI governance framework for a '''[team type/project type]''' that does '''[describe the work]'''. Map out all the places where team members might use AI in their workflow. For each touchpoint, classify the risk level:

- '''Low risk:''' AI errors are easily caught and consequences are minor (e.g., drafting internal emails, brainstorming)
- '''Medium risk:''' AI errors could waste significant time or create confusion (e.g., research summaries, data analysis, first drafts of client deliverables)
- '''High risk:''' AI errors could cause reputational, legal, or financial harm (e.g., published content, financial recommendations, legal language, customer-facing decisions)

Present this as a table with: Touchpoint | Description | Risk Level | Why
</blockquote>

'''Step 2 — Design the verification tiers.''' Based on the risk map, create a tiered verification system:

<blockquote>
Based on the risk map above, design a 3-tier verification system:

'''Tier 1 (Low risk):''' What's the minimum verification needed? What can proceed without review?
'''Tier 2 (Medium risk):''' What checks are required? Who reviews? What's the turnaround expectation?
'''Tier 3 (High risk):''' What's the full review process? Who signs off? What documentation is needed?

For each tier, specify:
- Verification steps (checklist)
- Who is responsible
- What "approved" looks like
- What happens when issues are found
</blockquote>

'''Step 3 — Write the team guidelines.''' Now produce the actual document:

<blockquote>
Write a 1-page "AI Usage Guidelines" document for this team. It should be practical, not corporate. Include:

1. '''When to use AI''' — Green light scenarios
2. '''When to be careful''' — Yellow light scenarios with required verification
3. '''When NOT to use AI''' — Red light scenarios or scenarios requiring explicit approval
4. '''Verification standards''' — The tier system from above, simplified
5. '''Attribution''' — When and how to disclose AI usage
6. '''Escalation''' — What to do when you're unsure whether AI use is appropriate

Write it in the tone of a senior colleague giving practical advice, not a legal department issuing mandates.
</blockquote>

'''Step 4 — Red-team the framework.''' Test it:

<blockquote>
Now role-play as a team member who wants to use AI in a gray area. Come up with 3 realistic scenarios where the guidelines are ambiguous or where a reasonable person might interpret them differently. For each scenario, suggest how to clarify the guideline.
</blockquote>

Revise the guidelines based on the edge cases.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose your context''' — Pick a real team or project. The framework should be one you could actually share or implement.
# '''Map AI touchpoints and risk levels''' — Identify every place AI could be used in the workflow and classify each by potential harm from errors.
# '''Design tiered verification''' — Create different verification processes for different risk levels. Not everything needs the same scrutiny.
# '''Write the guidelines''' — Produce a practical 1-page document that a team member could reference in their daily work.
# '''Red-team with edge cases''' — Test the framework against ambiguous scenarios. Fix any gaps before sharing.

'''"Done" looks like:''' A complete, practical AI governance framework (risk map + tiered verification + 1-page guidelines) that you could present to your team, plus documentation of edge cases you tested against.

----

== 🧭 Why this matters (Strategists start here) ==

Individual verification habits (from [[The Verification Checklist|EP-Intermediate-01]]) don't scale to teams. When five people use AI differently with different standards, the team's AI output quality is only as good as the weakest link. A governance framework creates '''shared standards without bureaucracy''' — it tells people what's safe to do quickly and what requires care, without making every AI interaction feel like a compliance exercise. This is also the document organizations will pay for: a practical, calibrated AI usage policy that actually gets followed.

----

== Reflection ==
* Which risk classification was hardest to assign? What does that ambiguity tell you?
* Would your team actually follow these guidelines? What would make them ignore it?
* Did the red-teaming step reveal fundamental gaps, or just edge cases?
* 💬 ''Present your framework to a colleague and ask: "Would you follow this?" Their honest reaction is more useful than any AI review.'' (Social Learners)

== ⬆️ Level up ==

You've reached the advanced level for Ethical Prompting & Judgment. From here, consider:
* Presenting this framework to your actual team and iterating based on feedback
* Combining this with [[Design Your Agent Workflow|AC-Advanced-01]] to add governance to multi-agent workflows
* Building a case study of how the framework changed AI usage behavior in your team

Back to [[Pillars/Ethical Prompting|Ethical Prompting & Judgment]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Ethical Prompting Exercises]]

Prompt Engineering Basics

2026-03-16T16:28:08Z

Admin: Fix 3 internal link(s)

''How to communicate with AI effectively. Not magic incantations — just clear thinking translated into clear instructions.''

<blockquote>
'''Plain English:''' Prompt engineering is the skill of giving AI clear, structured instructions that get useful results. It's not about memorizing "magic prompts" — it's about clear thinking. If you can write a good brief for a colleague, you can write a good prompt.
</blockquote>

== Why this matters ==

Here's a pattern you may recognize: you type something into an AI tool, get a mediocre response, and think "AI isn't that useful." But the problem usually isn't the AI — it's the prompt.

A vague prompt gives AI very little to work with. "Write something about marketing" could go in a thousand directions. "Write a 200-word email to my team explaining why we're shifting our Q2 campaign focus from brand awareness to lead generation, using a direct but supportive tone" gives the model enough context to produce something genuinely useful.

The difference between mediocre and excellent AI output is almost always in the prompt. And the good news is that the underlying techniques are simple to learn.

== The core techniques ==

=== Be specific about what you want ===

This is the most impactful improvement most people can make. Compare:

| Vague prompt || Specific prompt

| "Summarize this document" || "Summarize this document in 3 bullet points, focusing on budget implications for our team"
| "Write a response to this email" || "Write a professional but warm reply declining the meeting request, suggesting an async alternative"
| "Help me with my presentation" || "Give me 5 compelling opening lines for a presentation about remote work to an audience of skeptical middle managers"

You're not writing code. You're writing a brief — the same way you'd brief a colleague who's smart but doesn't know your context.

=== Give AI a role (system prompts) ===

Telling AI ''who it is'' changes the quality of its output dramatically. This is the foundation of the [[Your First AI Team Meeting|Your First AI Team Meeting]] exercise.

``<code>
You are a senior financial analyst reviewing a startup's pitch deck.
Focus on: revenue model assumptions, burn rate, and market size claims.
Be skeptical but constructive. Flag anything that seems unrealistic.
</code>`<code>

Why this works: it constrains the model's vast knowledge to a specific perspective and expertise level. Without a role, AI defaults to generic helpfulness. With a role, it brings a point of view.

=== Show examples (few-shot prompting) ===

If you want AI to produce output in a specific format or style, show it what good looks like:

</code>`<code>
Classify each customer comment as POSITIVE, NEGATIVE, or NEUTRAL.

Comment: "Love the new dashboard, it's so much faster"
Classification: POSITIVE

Comment: "The update broke my saved filters"
Classification: NEGATIVE

Comment: "I noticed you changed the sidebar layout"
Classification: NEUTRAL

Comment: "Can't believe you removed the export feature, this is useless now"
Classification:
</code>`<code>

Two or three examples are usually enough. The AI picks up the pattern — format, tone, judgment criteria — from your examples rather than having to guess what you mean.

=== Ask for step-by-step reasoning ===

For complex questions, adding "think step by step" or "show your reasoning" dramatically improves accuracy. This is called chain-of-thought prompting:

</code>`<code>
A company has 150 employees. They're cutting 12% of staff across
three departments: Engineering (80 people), Sales (45 people),
and Operations (25 people). The cuts should be proportional
to department size.

Think step by step: how many people are cut from each department?
</code>`<code>

Without the step-by-step instruction, AI often jumps to a wrong answer on multi-step problems. With it, the model works through the math visibly, and you can check each step.

=== Structure your prompt clearly ===

When your prompt has multiple parts — context, instructions, content to process — use clear structure to separate them:

</code>`<code>
== Context ==
I'm a project manager preparing for a stakeholder review meeting tomorrow.

== The document ==
[paste document here]

== What I need ==
# Three key risks I should be prepared to discuss
# One piece of good news I can lead with
# Any numbers that seem inconsistent and should be double-checked
</code>`<code>

This matters because AI processes your entire prompt as one stream of text. Without clear separation, it might confuse your instructions with the content you're asking it to analyze. Delimiters (like headers, brackets, or triple backticks) prevent this.

== Common mistakes ==

| Mistake || Why it fails || Better approach

| "Write something good about X" || Too vague — "good" means nothing to a model || Specify length, audience, tone, and purpose
| Putting instructions ''after'' the content || AI may lose track of late instructions in long prompts || Instructions first, content second
| Asking for everything at once || Quality drops when the task is too broad || Break complex tasks into smaller prompts, chain the outputs
| No examples for ambiguous tasks || The model interprets differently than you expect || Add 2–3 examples of what you want
| Accepting the first output || First drafts are rarely best || Ask for revisions: "Make this more concise" or "Rewrite focusing on X"

== Temperature: creativity vs. precision ==

Most AI tools let you adjust "temperature" — how creative vs. predictable the output is. You may not always have direct access to this setting, but understanding it helps:

| Setting || Behavior || Good for

| Low (precise) || Picks the most predictable words || Factual Q&A, classification, code, data extraction
| Medium || Balanced || General writing, summarization, analysis
| High (creative) || More varied and surprising || Brainstorming, creative writing, generating diverse options

Temperature doesn't change what the AI knows — it changes how it samples from possibilities. Low temperature = safe, predictable choices. High temperature = more variety, more risk of weirdness.

== How this connects to the playbook ==

Every exercise in this playbook is, at some level, a prompt engineering exercise. Here's how the techniques map:

| Technique || You'll practice it in

| '''Giving AI a role''' || [[Exercises/Agent Collaboration/Ac Basic 01 || Your First AI Team Meeting]] — dual expert roles
| '''Being specific''' || [[Exercises/Workflow Automation/Wa Basic 01 || The Reusable Prompt]] — building prompts worth keeping
| '''Showing examples''' || [[Exercises/Workflow Automation/Wa Intermediate 01 || The Prompt Chain]] — structured multi-step outputs
| '''Step-by-step reasoning''' || [[Exercises/Insight Synthesis/Is Basic 01 || The Signal in the Noise]] — extracting structured insight
| '''Clear structure''' || [[Exercises/Cross Domain Reframing/Cdr Intermediate 01 || The Framework Transplant]] — complex reframing prompts
| '''Iterating on output''' || [[Exercises/Ethical Prompting/Ep Basic 01 || The Fact-Check Habit]] — pushing back on AI responses

== The hierarchy of AI improvement ==

Before you look for a fancier tool, try a better prompt. This is the cheapest, fastest way to improve your AI output:

</code>`<code>
Most effort: Train a custom model
Fine-tune an existing model
RAG (give AI access to your documents)
Better prompts ← start here
Least effort: Use defaults and hope for the best
</code>``

Most people are somewhere between "use defaults" and "better prompts." Moving up just one level — from vague prompts to structured, specific prompts with roles and examples — is where the biggest improvement happens.

== Where to go next ==
* [[The Reusable Prompt|The Reusable Prompt]] — build a prompt you'll actually use again
* [[Your First AI Team Meeting|Your First AI Team Meeting]] — practice role-based prompting

[[Category:AI Fluency Playbook]]
[[Category:Core Concepts]]

Pathways

2026-03-16T16:28:07Z

Admin: Fix 5 internal link(s)

''Guided learning routes based on your AI Skills Quiz results. Each pathway targets specific strengths and gaps.''

Pathways are guided routes through the playbook. Each one is designed for a specific profile — a combination of strengths and growth areas. Find the one that fits you best.

== Available Pathways ==
* [[Pathway: Starting from Scratch|Starting from Scratch]] — New to AI? Start here. A gentle, step-by-step introduction.
* [[Pathway: Strong Communicator, Building Technical Confidence|Strong Communicator, Building Technical Confidence]] — Great with words and people, ready to build technical AI skills.
* [[Pathway: High Synthesis, Low Agent Collaboration|High Synthesis, Low Agent Collaboration]] — Strong at insight extraction, new to agent-based work.
* [[Pathway: High Automation, Low Ethics|High Automation, Low Ethics]] — Comfortable with automation, needs verification habits.

== Don't See Your Profile? ==

Start with the pillar where you scored lowest — that's where you'll get the most growth for the least effort. If you haven't taken the quiz, try the [https://aiskillsquiz.com AI Skills Quiz] to discover your profile.

You can also browse all [[Exercises: Pick Your Challenge|exercises]] and pick whatever catches your interest. There's no wrong way to navigate this playbook.

[[Category:AI Fluency Playbook]]
[[Category:Learning Pathways]]

Pathway: Strong Communicator, Building Technical Confidence

2026-03-16T16:28:06Z

Admin: Fix 6 internal link(s)

''A pathway for strong communicators building technical AI confidence alongside their existing strengths.''

== Your Profile ==

You're great with words, people, and ideas. You might be a writer, a manager, a consultant, a teacher, or someone who regularly translates complex things into clear language. Your Ethical Prompting and Insight Synthesis scores are likely your strongest pillars.

But when it comes to the more "technical" side of AI — building workflows, designing processes, working with agents — you feel less confident. That's not a weakness; it's just an area you haven't explored yet. And your communication skills are actually a superpower here.

== Recommended Sequence ==
# Start with [[The Signal in the Noise|The Signal in the Noise]] — This plays to your strength. You already know how to extract meaning from information; this exercise shows you how to do it systematically with AI.
# Then [[The Reusable Prompt|The Reusable Prompt]] — This is the bridge from "using AI" to "building with AI." It's more accessible than it sounds — if you can write a good email template, you can write a good prompt template.
# Then [[The Prompt Chain|The Prompt Chain]] — Build a 3-step AI pipeline. This is where the "technical" side starts to feel natural, because you're essentially designing a conversation flow — something you already do well.
# Then [[Your First AI Team Meeting|Your First AI Team Meeting]] — Use your facilitation instincts to orchestrate multiple AI perspectives. Communicators often excel at this exercise because it's fundamentally about managing different viewpoints.
# Stretch: [[The Multi-Source Brief|The Multi-Source Brief]] — Combine your synthesis and communication skills to triangulate AI perspectives into a clear brief.

== Common Pitfalls ==
* '''Thinking "technical" means "hard."''' Workflow automation and agent collaboration sound technical, but at their core they're about designing clear processes and managing information flow — things you already do in your work. The vocabulary is new; the underlying skills are not.
* '''Over-relying on your prompting strength.''' Because you're good with language, you can usually get decent results from AI through clever prompting alone. This can become a crutch that prevents you from learning to build repeatable systems. The goal is to move from "I can always write a good prompt" to "I've built a process that works even on autopilot."
* '''Avoiding exercises that feel "too technical."''' The Workflow Automation exercises might look intimidating, but [[The Reusable Prompt|The Reusable Prompt]] is literally just writing a template. Start there and you'll see that the "technical" pillar is more accessible than you expected.
* '''Undervaluing your existing skills.''' You might feel behind compared to people who are building AI automations and agent workflows. But your ability to think critically, communicate clearly, and synthesize information is foundational to AI fluency. The technical skills build on top of what you already have.

== What Leveling Up Looks Like ==
* You've built at least one multi-step AI workflow (a prompt chain or pipeline) and it works reliably
* You can design a simple AI-assisted process from scratch, not just use AI for one-off questions
* You're comfortable orchestrating multiple AI sessions for different parts of a project
* You've combined your communication strengths with new technical skills — for example, designing a workflow that ends with a polished, human-quality output
* When someone describes a repetitive task, your instinct is "I could build a prompt chain for that"

[[Category:AI Fluency Playbook]]
[[Category:Learning Pathways]]

Pathway: Starting from Scratch

2026-03-16T16:28:05Z

Admin: Fix 5 internal link(s)

''A guided pathway for people new to AI — building foundational skills from the ground up.''

== Your Profile ==

You're new to AI — or at least new to using it intentionally. Maybe you've played around with ChatGPT or Claude a few times, but you don't yet have a clear sense of what AI is good at, where it falls short, or how to make it genuinely useful in your work.

That's a perfectly good place to start. In fact, it's a great one — you don't have any bad habits to unlearn.

== Recommended Sequence ==
# Start with [[The Fact-Check Habit|The Fact-Check Habit]] — This is the single most important skill in AI fluency: learning to verify what AI tells you. It's quick, eye-opening, and you'll use it every time you work with AI from now on.
# Then [[The Signal in the Noise|The Signal in the Noise]] — Learn to pull useful information out of verbose AI responses. This makes every future AI interaction more productive.
# Then [[The Reusable Prompt|The Reusable Prompt]] — Turn something you do regularly into a prompt template. This is where AI starts saving you real time.
# Then branch out: [[The Stolen Technique|The Stolen Technique]] — Try borrowing a technique from a different field. This exercise expands how you think about AI's possibilities.
# Stretch: [[Your First AI Team Meeting|Your First AI Team Meeting]] — Once you're comfortable with the basics, try working with AI in multiple roles. This is a glimpse of where AI is heading.

== Common Pitfalls ==
* '''Trying to learn everything at once.''' There's a lot of AI content out there and it can feel like you need to absorb all of it. You don't. One exercise per week is enough. Depth beats breadth, especially at the beginning.
* '''Comparing yourself to power users.''' People who've been using AI tools for years had their own slow start. Your pace is fine. The fact that you're being intentional about learning puts you ahead of most people who just use AI casually.
* '''Dismissing AI too quickly when it gives a bad answer.''' Early on, you might conclude "AI isn't useful" after a disappointing response. Before you give up, try rephrasing your prompt or giving AI more context. The difference between a bad prompt and a good one is often just a few extra sentences.
* '''Skipping the reflection.''' It's tempting to rush through the exercises and move on. But the moment you pause to ask "what did I actually learn here?" is where the real skill development happens.

== What Leveling Up Looks Like ==
* You can explain to someone else what AI is good at and where it tends to fail
* You have a personal habit of verifying AI output before trusting it
* You've built at least one prompt template that you actually reuse in your work
* You feel confident enough to try an Intermediate exercise
* When someone asks "can AI do X?", you can give a thoughtful, nuanced answer instead of a yes or no

[[Category:AI Fluency Playbook]]
[[Category:Learning Pathways]]

Pathway: High Synthesis, Low Agent Collaboration

2026-03-16T16:28:05Z

Admin: Fix 5 internal link(s)

''A pathway for people strong at extracting AI insight but new to multi-agent collaboration.''

== Your Profile ==

You're strong at extracting meaning from AI output but haven't explored working with AI agents or multi-agent systems. This is the most common profile in the community.

== Recommended Sequence ==
# Start with [[Your First AI Team Meeting|AC-Basic-01]] — Get your feet wet with agent-based thinking
# Then [[The Handoff Protocol|AC-Intermediate-01]] — Apply your synthesis skills to multi-agent output
# Then [[The Multi-Source Brief|IS-Intermediate-01]] — Level up your strongest pillar
# Stretch: [[Design Your Agent Workflow|AC-Advanced-01]] — Design an agent workflow

== Common Pitfalls ==
* '''Treating agent collaboration as "just multi-prompting."''' You're good at getting insights from AI, so you may assume agent collaboration is just more of the same. It's not — it's about designing roles, managing handoffs, and building systems. The mental shift from "getting answers" to "orchestrating agents" is the hard part.
* '''Over-synthesizing, under-building.''' Your strength in synthesis can keep you in analysis mode — reading, comparing, evaluating — without moving to building actual workflows. At some point, you need to design and run a multi-agent process, not just think about it.
* '''Skipping the basic agent exercise.''' If you score high on Insight Synthesis, you may feel that [[Your First AI Team Meeting|AC-Basic-01]] is beneath you. It's not. The exercise introduces a mental model (role-based AI interaction) that's fundamentally different from single-query synthesis work. Don't skip the foundation.
* '''Applying old patterns to new territory.''' You may try to use your synthesis skills (asking AI good questions, evaluating output quality) as a substitute for agent collaboration skills (defining roles, managing context boundaries, designing handoffs). Both matter, but they're different muscles.

== What Leveling Up Looks Like ==
* You can split a complex task across multiple AI sessions with different roles and produce output that no single session could have generated
* You design agent workflows before running them — mapping roles, inputs, outputs, and handoffs on paper first
* You naturally think about context boundaries: what each AI session should and shouldn't know
* You combine your synthesis skills with agent collaboration: using your ability to evaluate and integrate output as the orchestration layer between specialized AI agents
* When facing a complex problem, your instinct shifts from "let me ask AI about this" to "let me design a multi-perspective approach"

[[Category:AI Fluency Playbook]]
[[Category:Learning Pathways]]

Pathway: High Automation, Low Ethics

2026-03-16T16:28:04Z

Admin: Fix 5 internal link(s)

''A pathway for automation-skilled users who need to build stronger verification and ethical judgment habits.''

== Your Profile ==

You're comfortable building AI-assisted workflows but may be moving fast without checking outputs carefully. This pathway helps you build verification habits into your existing automation skills.

== Recommended Sequence ==
# Start with [[The Fact-Check Habit|EP-Basic-01]] — Establish a baseline for checking AI output
# Then [[The Verification Checklist|EP-Intermediate-01]] — Build verification into your workflows
# Then [[The Prompt Chain|WA-Intermediate-01]] — Level up automation with ethics baked in
# Stretch: [[The AI Governance Playbook|EP-Advanced-01]] — Design an accountability system

== Common Pitfalls ==
* '''Seeing verification as a speed bump.''' Your automation instinct treats every additional step as overhead. But verification isn't overhead — it's quality assurance. The goal is to make verification ''efficient'', not to eliminate it.
* '''Building trust in AI consistency.''' Because your workflows run smoothly most of the time, you develop false confidence that AI output is reliable. AI fails unpredictably — the 99 correct outputs make the 1 confidently wrong output much more dangerous.
* '''Confusing "I automate ethically" with "I have ethical habits."''' You may believe your intentions are good (and they probably are), but good intentions without systematic checks produce the same result as no intentions at all. The question isn't whether you care about accuracy — it's whether your processes guarantee it.
* '''Resisting the checklist.''' If you're the type who builds systems, using someone else's checklist feels awkward. But [[The Verification Checklist|EP-Intermediate-01]] asks you to build ''your own'' — customized to your work, tested against real outputs, and integrated into your existing workflows.

== What Leveling Up Looks Like ==
* Every AI workflow you build includes an explicit quality gate — a point where output is checked before it moves forward
* You can classify your AI-assisted tasks by risk level and apply appropriate verification standards (not everything gets the same scrutiny — but nothing gets zero scrutiny)
* You catch AI errors ''before'' they reach stakeholders, not after
* You've built a personal verification checklist and it's become habit — you use it without thinking about it
* When you design a new automation, your default question is "where should the human check go?" rather than "how do I eliminate all human steps?"

[[Category:AI Fluency Playbook]]
[[Category:Learning Pathways]]

How to Use This Playbook

2026-03-16T16:28:03Z

Admin: Fix 8 internal link(s)

''Three ways to navigate the AI Fluency Playbook, plus tips for getting the most out of it.''

There's no single "right" way to use this playbook. It's designed to meet you where you are. Here are three entry paths — pick the one that resonates.

== Path 1: "I want to try something right now" ==

Perfect. Head to the [[Exercises: Pick Your Challenge|Exercises]] page, pick any basic exercise (15 minutes each), and start. No prep needed — just you and an AI tool of your choice.

== Path 2: "I took the AI Skills Quiz" ==

If you've taken the quiz at [https://aiskillsquiz.com aiskillsquiz.com], you already know your pillar scores and archetype. Here's your next step:
# '''Find your archetype''' in the [[Archetypes: Your Learning Style|Archetypes]] section to understand your learning style
# '''Check the [[Pathways|Pathways]]''' for a guided route matched to your profile
# '''Start with your lowest-scoring pillar''' — that's where you'll see the most growth

== Path 3: "I want to understand the big picture" ==

Start with the framework:
# Read [[The Five Pillars of AI Fluency|The Five Pillars]] to understand what AI fluency is made of
# Browse the [[Exercises: Pick Your Challenge|Exercises]] to see the full range from Basic to Advanced
# Then pick any exercise that interests you

== How the Playbook Is Organized ==

``<code>
Pillars (5 skill areas)
-> Exercises (Basic, Intermediate, Advanced)

Archetypes (your learning style)
-> Pathways (guided routes based on your strengths and gaps)

Resources (glossary, tools, further reading)
</code>``

Each '''exercise''' includes:
* A clear goal and time estimate
* Step-by-step instructions
* Multiple entry points for different learning styles
* Reflection prompts to deepen your understanding

== How Much Time Does This Take? ==

'''15 minutes a week is enough to start.''' Each Basic exercise is designed to fit into a coffee break. As you get comfortable, Intermediate exercises take about 25 minutes, and Advanced exercises around 40 minutes.

There's no deadline. This isn't a course you need to finish — it's a resource you return to as your skills and questions grow.

== Learning Style Entry Points ==

Every exercise offers multiple ways in:
* '''Jump in''' — For [[The Tinkerer|Tinkerers]]. Start doing, then understand why.
* '''Plan first''' — For [[The Planner|Planners]]. Read the overview, then execute step by step.
* '''Why this matters''' — For [[The Strategist|Strategists]]. Understand the strategic value before starting.
* '''Discuss''' — For [[The Social Learner|Social Learners]]. Use the reflection prompts with a colleague or community.

You don't have to pick just one — but knowing your preference can help you get started faster.

== Tips for Getting the Most Out of This ==
# '''Do the exercises with real work.''' The more relevant the topic, the more you'll learn.
# '''Don't skip the reflection.''' The "what did I learn" moment is where growth actually happens.
# '''Come back.''' Revisiting an exercise after a few weeks often reveals new insights.
# '''Share what you learn.''' Teaching is the best way to solidify understanding.

Ready? Pick a path above, or head back to the [[home page]] to explore.

[[Category:AI Fluency Playbook]]

How AI Actually Works

2026-03-16T16:28:02Z

Admin: Fix 3 internal link(s)

''A no-jargon explanation of what happens when you type something into ChatGPT, Claude, or Gemini — and why it matters for using AI well.''

<blockquote>
'''Plain English:''' When you type a message to an AI, it doesn't "think" or "know" things. It predicts the most likely next words based on patterns it learned from enormous amounts of text. Understanding this one idea changes how you use it.
</blockquote>

== The simplest explanation that's still true ==

Here's what happens when you send a message to ChatGPT, Claude, Gemini, or any other AI chat tool:
# '''Your message gets broken into tokens.''' The AI doesn't read words the way you do. It splits your text into small chunks called [[Glossary|tokens]] — sometimes whole words, sometimes pieces of words. The word "tokenization" might become "token" + "ization." This is why AI tools have limits on how much text you can send — they're counting tokens, not words. (For a deeper look at this, see [[Tokenization & Context Windows|Tokenization & Context Windows]].)
# '''Each token gets converted into numbers.''' The AI represents every token as a list of numbers (called an embedding) that captures its meaning and relationships. Words with similar meanings end up with similar numbers — "happy" and "joyful" are close together, "happy" and "wrench" are far apart.
# '''The model predicts what comes next.''' This is the core of it. The AI looks at all the tokens in your message, weighs how they relate to each other (this is the "attention" mechanism you might have heard about), and predicts the most likely next token. Then it predicts the next one. Then the next. One token at a time, until it has a complete response.

That's it. No understanding. No reasoning in the human sense. Pattern matching at a scale and sophistication that produces remarkably useful output — but pattern matching nonetheless.

== Why "trained on text" matters ==

Modern AI models like GPT-4, Claude, and Gemini were trained on enormous amounts of text — books, websites, code, conversations, research papers. During training, the model was repeatedly shown text with a word removed and asked to predict what goes there. Billions of times. Across billions of examples.

This is why AI can write in any style, answer questions about almost any topic, and generate code in dozens of languages. It's seen patterns in all of those domains.

But it also means:
* '''AI doesn't "know" facts.''' It learned that certain words tend to follow other words in certain contexts. When it tells you that Paris is the capital of France, it's not retrieving a fact from a database — it's producing the statistically likely continuation of your question. This is also why it can confidently state things that are wrong. (See: [[Why AI Gets Things Wrong|Why AI Gets Things Wrong]])
* '''Training has a cutoff date.''' The model learned from text up to a certain point. It doesn't know what happened after that unless it has access to search tools or uploaded documents.
* '''It reflects the patterns in its training data.''' Including the biases, the common phrasings, and the popular opinions. AI doesn't have its own perspective — it has a statistical average of human text.

== The three types you'll encounter ==

You don't need to memorize the full AI taxonomy, but understanding three distinctions helps:

'''AI''' is the broadest term — any system that performs tasks typically requiring human intelligence. Your email spam filter is AI. Siri is AI. A chess engine is AI.

'''Machine learning''' is a subset — systems that learn from data rather than following pre-written rules. Instead of programming "if the email contains 'Nigerian prince,' mark as spam," you show the system thousands of spam and non-spam emails and let it figure out the patterns.

'''Large Language Models (LLMs)''' are what you interact with when you use ChatGPT, Claude, or Gemini. They're a specific type of deep learning model trained on text, using an architecture called a "transformer." When people say "AI" in 2026, they usually mean this.

For the purposes of this playbook, when we say "AI," we mean the tools you actually use — the chat interfaces, the built-in AI features in your apps, the models you can give instructions to. The engineering details are fascinating (and if you want them, [https://xuecodex.tech/docs XueCodex] goes deep), but you don't need them to build AI fluency.

== What this means for how you use AI ==

Understanding that AI predicts rather than knows changes your approach in practical ways:

'''Give it more context, not less.''' The model predicts based on what's in front of it. A vague prompt gives it less to work with, so predictions are more generic. A detailed prompt with context, examples, and constraints gives it much better patterns to follow. This is why [[Prompt Engineering Basics|prompt engineering]] matters.

'''Verify, don't trust.''' Since the model is generating plausible text rather than retrieving verified facts, you need to check outputs — especially for specific claims, numbers, dates, and citations. The [[The Fact-Check Habit|Fact-Check Habit]] exercise builds this skill.

'''It's a collaborator, not an oracle.''' AI is most useful when you bring judgment and it brings speed and breadth. You know your context, your goals, your stakeholders. It can process information, generate options, and stress-test your thinking faster than you can alone.

'''Different models have different strengths.''' GPT, Claude, and Gemini are trained differently, on different data, with different priorities. A prompt that works brilliantly in one may flop in another. This is why the playbook is tool-agnostic — we teach capabilities, not product-specific tricks.

== Where to go next ==
* [[Why AI Gets Things Wrong|Why AI Gets Things Wrong]] — what hallucinations are and why they happen
* [[The Fact-Check Habit|The Fact-Check Habit]] — your first exercise in working with AI critically

[[Category:AI Fluency Playbook]]
[[Category:Core Concepts]]

Getting Started: What Is AI Fluency?

2026-03-16T16:28:01Z

Admin: Fix 7 internal link(s)

''What AI fluency means and why it matters for generalists.''

You don't need to be a developer to be great with AI.

AI fluency isn't about writing code, mastering every tool, or knowing the right "magic prompts." It's about learning to think clearly alongside AI — knowing when to trust it, when to push back, and how to make it genuinely useful in your work. If you want to understand the full picture of what AI fluency means and why it matters for generalists, see [[What We Mean by AI Fluency|What We Mean by AI Fluency]].

This playbook was built for '''generalists''' — people who wear many hats, work across domains, and don't fit neatly into a single job title. No technical background needed. If you can have a conversation with ChatGPT, Claude, or similar tools, you have everything you need to start.

== What You'll Find Here ==
* '''15 hands-on exercises''' you can do in 15-40 minutes, no setup required
* '''[[The Five Pillars of AI Fluency|Five skill pillars]]''' that break AI fluency into concrete, buildable areas
* '''[[Archetypes: Your Learning Style|Learning style archetypes]] and [[Pathways|pathways]]''' that personalize your journey
* '''[[Resources|Resources]]''' including a glossary, tool recommendations, and further reading

== Ready to Start? ==

'''Take the quiz.''' The [https://aiskillsquiz.com AI Skills Quiz] takes a few minutes and gives you a personalized profile — including your strengths, gaps, and a recommended pathway.

'''Or jump straight into an exercise.''' Browse the full [[Exercises: Pick Your Challenge|exercise list]] or start with [[The Fact-Check Habit|The Fact-Check Habit]] — it takes 15 minutes and you'll learn something immediately.

'''Or read the big picture first.''' Check out [[How to Use This Playbook|How to Use This Playbook]] for three navigation paths, or [[The Five Pillars of AI Fluency|The Five Pillars]] to understand the full framework.

[[Category:AI Fluency Playbook]]

Exercises: Pick Your Challenge

2026-03-16T16:28:00Z

Admin: Fix 24 internal link(s)

''15 hands-on AI exercises across five skill pillars and three difficulty levels. Pick any exercise and start building practical AI fluency.''

Every exercise in this playbook is designed to be self-contained — you can do any of them with just an AI tool and a bit of curiosity. No setup, no prerequisites, no special software.

== Not Sure Where to Start? ==

'''Got 15 minutes?''' Try one of these beginner-friendly exercises:
* [[The Fact-Check Habit|The Fact-Check Habit]] — Catch an AI making a mistake (you'll be surprised how easy it is)
* [[The Signal in the Noise|The Signal in the Noise]] — Turn a messy AI brainstorm into something useful
* [[The Reusable Prompt|The Reusable Prompt]] — Build a prompt template you'll actually use again

'''Got 25 minutes?''' Level up with an intermediate challenge:
* [[The Prompt Chain|The Prompt Chain]] — Build a 3-step AI pipeline
* [[The Multi-Source Brief|The Multi-Source Brief]] — Triangulate multiple AI perspectives
* [[The Verification Checklist|The Verification Checklist]] — Create your personal AI fact-checking system

'''Ready for a deep dive?''' Advanced exercises take about 40 minutes:
* [[The Workflow Blueprint|The Workflow Blueprint]] — Design a complete AI-assisted workflow
* [[Design Your Agent Workflow|Design Your Agent Workflow]] — Architect a multi-agent system
* [[The AI Governance Playbook|The AI Governance Playbook]] — Build team-level AI guidelines

== All Exercises by Pillar ==

| Pillar || Basic (15 min) || Intermediate (25 min) || Advanced (40 min)

| Ethical Prompting || [[Exercises/Ethical Prompting/Ep Basic 01 || Fact-Check Habit]] || [[Exercises/Ethical Prompting/Ep Intermediate 01 || Verification Checklist]] || [[Exercises/Ethical Prompting/Ep Advanced 01 || AI Governance Playbook]]
| Insight Synthesis || [[Exercises/Insight Synthesis/Is Basic 01 || Signal in the Noise]] || [[Exercises/Insight Synthesis/Is Intermediate 01 || Multi-Source Brief]] || [[Exercises/Insight Synthesis/Is Advanced 01 || Research Pipeline]]
| Workflow Automation || [[Exercises/Workflow Automation/Wa Basic 01 || Reusable Prompt]] || [[Exercises/Workflow Automation/Wa Intermediate 01 || Prompt Chain]] || [[Exercises/Workflow Automation/Wa Advanced 01 || Workflow Blueprint]]
| Cross-Domain Reframing || [[Exercises/Cross Domain Reframing/Cdr Basic 01 || Stolen Technique]] || [[Exercises/Cross Domain Reframing/Cdr Intermediate 01 || Framework Transplant]] || [[Exercises/Cross Domain Reframing/Cdr Advanced 01 || Cross-Domain Library]]
| Agent Collaboration || [[Exercises/Agent Collaboration/Ac Basic 01 || First AI Team Meeting]] || [[Exercises/Agent Collaboration/Ac Intermediate 01 || Handoff Protocol]] || [[Exercises/Agent Collaboration/Ac Advanced 01 || Agent Workflow Design]]

== All Exercises by Level ==

=== Basic (15 min each) ===
Foundational skills and first interactions with AI. Start here if you're new.
* [[The Fact-Check Habit|The Fact-Check Habit]] — Ethical Prompting
* [[The Signal in the Noise|The Signal in the Noise]] — Insight Synthesis
* [[The Reusable Prompt|The Reusable Prompt]] — Workflow Automation
* [[The Stolen Technique|The Stolen Technique]] — Cross-Domain Reframing
* [[Your First AI Team Meeting|Your First AI Team Meeting]] — Agent Collaboration

=== Intermediate (25 min each) ===
Applied skills and workflow integration. You're chaining steps and building processes.
* [[The Verification Checklist|The Verification Checklist]] — Ethical Prompting
* [[The Multi-Source Brief|The Multi-Source Brief]] — Insight Synthesis
* [[The Prompt Chain|The Prompt Chain]] — Workflow Automation
* [[The Framework Transplant|The Framework Transplant]] — Cross-Domain Reframing
* [[The Handoff Protocol|The Handoff Protocol]] — Agent Collaboration

=== Advanced (40 min each) ===
Strategic use and system-level thinking. You're designing frameworks for teams.
* [[The AI Governance Playbook|The AI Governance Playbook]] — Ethical Prompting
* [[The Research Pipeline|The Research Pipeline]] — Insight Synthesis
* [[The Workflow Blueprint|The Workflow Blueprint]] — Workflow Automation
* [[The Cross-Domain Prompt Library|The Cross-Domain Prompt Library]] — Cross-Domain Reframing
* [[Design Your Agent Workflow|Design Your Agent Workflow]] — Agent Collaboration

You don't need to do them in order. Pick whatever catches your interest — the best exercise is the one you actually do.

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]

Design Your Agent Workflow

2026-03-16T16:27:58Z

Admin: Fix 1 internal link(s)

''Architect a complete multi-agent workflow with defined roles, handoffs, and feedback loops. 40 minutes.''

<blockquote>
'''One-liner:''' Architect a complete multi-agent workflow for a real project — defining roles, inputs, outputs, handoffs, and a feedback loop — then test it.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a real project that involves at least three distinct types of work (research, analysis, creation, review, etc.). Examples: writing a report, planning an event, developing a proposal, building a content calendar.

'''Design a 3-4 agent workflow on paper or in a doc.''' For each agent, define:

| Agent Role || What it receives (input) || What it produces (output) || Handoff trigger

| Agent 1: Researcher || The project brief || A structured summary of key findings || "Research complete" + summary ready
| Agent 2: Drafter || Research summary + project brief || A first draft || Draft complete
| Agent 3: Critic || The draft + original brief || Specific critique with improvement suggestions || Review complete
| Agent 4: Editor || Draft + critique notes || Final polished output || Revisions applied

Now '''implement it''' using chained AI prompts. Open a chat for each agent (or reuse one chat with fresh role prompts). Run the workflow end-to-end:

'''Agent 1 prompt:'''
<blockquote>
You are a '''research analyst'''. Your job is to gather and organize relevant information. Here is the project brief: '''[paste your brief]'''. Produce a structured summary of the key information I'll need. Organize it by theme. Include 3-5 key insights and any risks or gaps you see.
</blockquote>

Take Agent 1's output and feed it to Agent 2:

'''Agent 2 prompt:'''
<blockquote>
You are a '''content drafter'''. Your job is to turn research into a clear first draft. Here is the project brief: '''[paste brief]'''. Here is the research summary: '''[paste Agent 1 output]'''. Write a first draft that addresses the brief. Focus on clarity and completeness. Don't self-edit — that's someone else's job.
</blockquote>

Take Agent 2's output and feed it to Agent 3:

'''Agent 3 prompt:'''
<blockquote>
You are a '''critical reviewer'''. Your job is to find weaknesses and suggest improvements. Here is the original brief: '''[paste brief]'''. Here is the draft: '''[paste Agent 2 output]'''. Identify: (1) gaps — what's missing that the brief requires, (2) weaknesses — arguments or sections that aren't convincing, (3) specific improvement suggestions with rationale. Do NOT rewrite the draft. Just critique.
</blockquote>

Take the draft and critique to Agent 4:

'''Agent 4 prompt:'''
<blockquote>
You are a '''senior editor'''. Your job is to produce the final version. Here is the draft: '''[paste Agent 2 output]'''. Here is the review feedback: '''[paste Agent 3 output]'''. Revise the draft to address the critique. Maintain the original structure where it works. Explain your key changes in a brief editor's note at the end.
</blockquote>

'''Feedback loop (optional):''' Take the final output and feed it back to Agent 3 for a second review. Notice how the quality changes with each iteration.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a project''' — Pick something with enough complexity to benefit from specialization. A single-paragraph task won't stretch this exercise. Good candidates: a report, a strategy document, a proposal, or a content plan.
# '''Design the agent architecture''' — Map out 3-4 agent roles using the table format above. Define clear inputs, outputs, and handoff triggers for each. The key design decision: what does each agent ''not'' know or ''not'' do?
# '''Write the role prompts''' — Create a system-level prompt for each agent that sets its role, scope, and constraints. Explicitly state what's out of scope for each agent.
# '''Run the workflow sequentially''' — Execute each agent in order, manually passing outputs between them. Track what you pass and what you leave out.
# '''Evaluate the result''' — Compare the final output to what you'd get from a single "do everything" prompt. Document what the workflow architecture added.

'''"Done" looks like:''' You have a documented agent workflow (the architecture) and a finished output that went through the full pipeline. You can explain why you split the work the way you did and what each agent contributed.

----

== 🧭 Why this matters (Strategists start here) ==

This is what agent collaboration looks like at professional scale — '''architecture before implementation'''. Every multi-agent framework (CrewAI, AutoGen, LangGraph) requires you to define roles, handoffs, and feedback loops before writing a single line of code. By doing it manually first, you understand the design decisions that make or break an agent system: what context each agent needs, where handoffs lose information, and when feedback loops help vs. when they add noise. This exercise builds the mental model that transfers to any agent tooling.

----

== Reflection ==
* Which agent in your workflow had the biggest impact on output quality? Would the workflow still work without the weakest agent?
* What information was lost between handoffs? Would you design the handoffs differently next time?
* Where did the feedback loop help, and where did it just add noise? Is there a point of diminishing returns?
* 💬 ''Walk a colleague through your agent architecture diagram before showing them the output. Ask them to predict where the pipeline would break — their predictions vs. reality reveals whether your architecture is intuitive or over-designed.'' (Social Learners)

== ⬆️ Level up ==

You've reached the advanced level for Agent Collaboration. From here, consider:
* Exploring agent frameworks like CrewAI or AutoGen to automate the handoffs you did manually
* Combining this skill with [[The Workflow Blueprint|WA-Advanced-01]] to build end-to-end automated workflows
* Revisiting this exercise with a more complex project to push the architecture further

Back to [[Agent Collaboration|Agent Collaboration]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Agent Collaboration Exercises]]

Core Concepts

2026-03-16T16:27:57Z

Admin: Fix 1 internal link(s)

''Foundational ideas that underpin the entire playbook — how AI works, why it fails, and how to communicate with it effectively.''

Before diving into exercises and pillars, it helps to understand a few foundational ideas. These concept pages give you the mental models you need to use AI well — no technical background required.
* [[What We Mean by AI Fluency|What We Mean by AI Fluency]] — What AI fluency actually is, why it matters for generalists, and how the five pillars fit together
* [[How AI Actually Works|How AI Actually Works]] — What happens when you type something into ChatGPT, Claude, or Gemini
* [[Tokenization & Context Windows|Tokenization & Context Windows]] — Why AI can only "remember" so much, and what you can do about it
* [[Why AI Gets Things Wrong|Why AI Gets Things Wrong]] — What hallucinations are, why they happen, and how to catch them
* [[Prompt Engineering Basics|Prompt Engineering Basics]] — How to communicate with AI effectively using clear, structured instructions
* [[Agents vs. Assistants|Agents vs. Assistants]] — The spectrum from simple chatbot to autonomous agent, and how to choose the right level

These pages are referenced throughout the playbook's exercises and pillars. You don't need to read them all before starting — but they're here when you want to understand the ''why'' behind the techniques.

[[Category:AI Fluency Playbook]]
[[Category:Core Concepts]]

Agents vs. Assistants

2026-03-16T16:27:56Z

Admin: Fix 4 internal link(s)

''Not all AI works the same way. Understanding the spectrum from simple chatbot to autonomous agent helps you pick the right approach for the job.''

<blockquote>
'''Plain English:''' An AI assistant waits for you to ask it something and responds. An AI agent can plan steps, use tools, and take actions — with varying degrees of independence. Most of what you use today is somewhere in between.
</blockquote>

== The spectrum ==

People talk about "AI agents" like it's a single thing. It's not. There's a spectrum of autonomy, and understanding where different tools sit on it helps you choose the right approach:

``<code>
You do everything AI does everything
↓ ↓
[Chatbot] → [Assistant] → [Copilot] → [Agent] → [Autonomous Agent]
</code>``

'''Chatbot''' — You ask, it answers. No memory, no tools, no planning. A basic ChatGPT conversation with no custom instructions. You drive everything.

'''Assistant''' — It responds to your requests but can also follow standing instructions. Claude with a system prompt, a custom GPT with specific behaviors configured. It has a personality and constraints, but still only acts when you ask.

'''Copilot''' — It works alongside you in real-time, proactively suggesting things. GitHub Copilot auto-completing your code, Notion AI offering to summarize your page, Gmail suggesting replies. It's watching your work and offering help without being asked.

'''Agent''' — It can plan a multi-step task, decide which tools to use, and execute steps on its own. You give it a goal ("research these three competitors and draft a comparison table"), and it figures out the steps: search the web, read several pages, extract key data, format the output. You review the result, not each step.

'''Autonomous agent''' — It operates with minimal human oversight over extended periods. It monitors, decides, and acts. These are still emerging and mostly experimental — think automated trading systems or self-healing infrastructure monitoring.

== Where you actually are in 2026 ==

Most generalists interact with AI in the assistant-to-copilot range. And that's fine — there's enormous value there that most people haven't fully tapped yet.

But agent-level tools are becoming accessible to non-developers:
* '''Claude Projects''' — persistent context that makes Claude act more like an assistant who knows your work, less like a blank chatbot
* '''Custom GPTs''' — pre-configured assistants with specific knowledge and instructions
* '''Claude with tool use / web search''' — the AI decides when to search the web, read a document, or run code, then does it
* '''Cowork / Claude Code''' — full agent capability: reads your files, plans multi-step tasks, executes them, asks for your input at key decision points
* '''MCP (Model Context Protocol)''' — a standard that lets AI connect to your other tools (calendar, email, databases), so it can act on real information rather than just chat about it

The progression from the [[Agent Collaboration|Agent Collaboration]] exercises mirrors this spectrum exactly:
* '''Basic''' ([[Your First AI Team Meeting|Your First AI Team Meeting]]) — giving AI roles in a conversation (assistant level)
* '''Intermediate''' ([[The Handoff Protocol|The Handoff Protocol]]) — coordinating between AI sessions (copilot level)
* '''Advanced''' ([[Design Your Agent Workflow|Design Your Agent Workflow]]) — designing multi-step AI workflows (agent level)

== The key question: how much autonomy should you give? ==

More autonomy isn't always better. The right level depends on three things:

'''Stakes.''' How bad is it if the AI gets it wrong? For brainstorming ideas, high autonomy is fine — a wrong suggestion costs nothing. For sending an email to a client, you want to review before it sends. For financial calculations, you verify every number.

'''Predictability.''' How well-defined is the task? Formatting a weekly report from the same data sources is highly predictable — good candidate for an agent. "Help me figure out our strategy for next quarter" requires judgment at every step — keep it as a collaborative conversation.

'''Your expertise.''' Can you evaluate the output? If you're an expert in the domain, you can give AI more autonomy because you'll catch mistakes quickly. If you're learning a new area, keep the AI in assistant mode where you're directing every step and building your own understanding.

A practical framework:

| Autonomy level || Use when || Watch out for

| '''You direct, AI executes''' || High stakes, new domains, learning || Slower, but you understand everything
| '''AI proposes, you approve''' || Medium stakes, familiar territory || Review carefully — don't rubber-stamp
| '''AI acts, you spot-check''' || Low stakes, predictable tasks, repeatable workflows || Set up verification checkpoints
| '''AI acts autonomously''' || Very low stakes, highly predictable, easily reversible || Only if you can undo mistakes cheaply

== Common misconceptions ==

'''"I need agents to be AI-fluent."''' No. Most of the value in AI fluency comes from being excellent at the assistant level — writing great prompts, giving AI useful roles, structuring your requests clearly. The [[Prompt Engineering Basics|Prompt Engineering Basics]] matter more than any agent framework.

'''"Agents will replace my job."''' Agents automate ''tasks'', not ''roles''. A marketing generalist who uses AI agents to automate report formatting, competitive research, and first-draft content isn't being replaced — they're spending more time on strategy, relationships, and creative judgment. The [[Workflow Automation|Workflow Automation]] pillar is built on this distinction.

'''"Multi-agent systems are the future."''' You'll hear a lot about "teams of AI agents" working together. This is real technology, but it's overhyped for most generalists in 2026. One well-configured AI assistant with good context will outperform a poorly designed multi-agent system. Master the fundamentals first.

'''"More tools = more capable."''' Connecting AI to every tool in your stack sounds powerful, but every connection is a potential failure point and a security consideration. Start with one integration that saves you real time, get comfortable with it, then expand. The [[Pillars/Ethical Prompting|Ethical Prompting]] pillar covers the judgment side of this.

== Where to go next ==
* [[Agent Collaboration|Agent Collaboration pillar]] — the full progression from basic to advanced
* [[Your First AI Team Meeting|Your First AI Team Meeting]] — start with role-based prompting (no tools needed)

[[Category:AI Fluency Playbook]]
[[Category:Core Concepts]]

MediaWiki:Sidebar

2026-03-16T16:27:55Z

Admin: Add playbook navigation sidebar

Main Page

2026-03-16T16:27:54Z

Admin: Fix links, table formatting, and navigation

__NOTOC__
<div style="border: 2px solid #5c0098; border-radius: 8px; padding: 16px; margin-bottom: 20px; background: #f9f5ff;">
'''AI Fluency Playbook''' — A practical guide to building AI skills, designed for generalists, by generalists.
</div>

AI is changing how we work, but most resources assume you're a developer or data scientist. This playbook is different. It's built for people who work across domains, wear multiple hats, and want to use AI thoughtfully and effectively — without needing a technical background.

Whether you're just getting started or looking to level up, there's something here for you.

== Start Here ==

* '''[[Getting Started: What Is AI Fluency?|Getting Started]]''' — What AI fluency means and why it matters
* '''[[How to Use This Playbook]]''' — Three paths in, based on how you like to learn

== Explore the Playbook ==

{| class="wikitable" style="width:100%;"
|-
! Section !! What You'll Find
|-
| '''[[The Five Pillars of AI Fluency|Pillars]]''' || The five core skill areas of AI fluency
|-
| '''[[Exercises: Pick Your Challenge|Exercises]]''' || 15 hands-on challenges across three levels, from 15-minute basics to deep dives
|-
| '''[[Archetypes: Your Learning Style|Archetypes]]''' || Discover your AI learning profile
|-
| '''[[Pathways]]''' || Guided routes based on your strengths and gaps
|-
| '''[[Resources]]''' || Glossary, tools, and further reading
|}

== Take the Quiz ==

Not sure where to start? The [https://aiskillsquiz.com AI Skills Quiz] takes a few minutes and gives you a personalized profile — including your archetype, pillar scores, and a recommended pathway.

== Quick Start ==

Just want to try something? These three exercises are great first picks:

# '''[[The Fact-Check Habit]]''' — Learn to catch AI mistakes (15 min)
# '''[[The Signal in the Noise]]''' — Turn messy output into clear insight (15 min)
# '''[[The Reusable Prompt]]''' — Build a prompt you'll actually reuse (15 min)

----

''This playbook is part of the [https://generalistworld.com Generalist World] community. It's open, evolving, and built on the belief that AI fluency is a skill anyone can develop.''

[[Category:AI Fluency Playbook]]

Your First AI Team Meeting

2026-03-16T16:23:23Z

Admin: Fix 1 internal link(s)

''Run a multi-perspective AI session with two expert viewpoints on the same problem. 15 minutes.''

<blockquote>
'''One-liner:''' Run a multi-perspective AI session where one prompt gets you two expert viewpoints on the same problem — no extra tools required.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a real decision you're currently facing. It could be a work decision, a project direction, or a problem you're stuck on.

Paste this prompt into any AI chat (ChatGPT, Claude, Gemini — anything works):

<blockquote>
I want you to act as two different experts giving me advice on '''[your problem here]'''.

First, respond as a '''[Role A]''' — someone who focuses on '''[their priority]'''.
Then, respond as a '''[Role B]''' — someone who focuses on '''[their different priority]'''.

Keep each perspective clearly labeled. Be specific and give concrete recommendations, not vague advice.
</blockquote>

'''Example — choosing whether to launch a feature now or wait:'''

<blockquote>
I want you to act as two different experts giving me advice on whether to launch our new onboarding flow this week or wait until next month.

First, respond as a '''growth-focused product manager''' — someone who prioritizes user acquisition and speed to market.
Then, respond as a '''risk-aware QA lead''' — someone who prioritizes stability, edge cases, and user trust.

Keep each perspective clearly labeled. Be specific and give concrete recommendations, not vague advice.
</blockquote>

After reading both perspectives, send this follow-up:

<blockquote>
Now, act as a '''neutral facilitator'''. Summarize where these two experts agree, where they disagree, and what the key trade-off is. End with a single question I should answer before making my decision.
</blockquote>

Read the synthesis. Notice how one prompt gave you a structured debate that would normally require two people in a room.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose your problem''' — Pick a real decision or challenge you're working on right now. It works best when reasonable people could disagree about the right approach.
# '''Pick two expert roles''' — Choose two perspectives that would naturally see your problem differently. Examples: marketer vs. engineer, short-term thinker vs. long-term strategist, customer advocate vs. operations manager.
# '''Write and send the dual-role prompt''' — Use the template in the "Jump in" section. Fill in your problem and your two roles.
# '''Read both perspectives''' — Notice where they conflict, where they agree, and which one you instinctively lean toward.
# '''Send the facilitator follow-up''' — Ask the AI to synthesize the two views and surface the core trade-off.

'''"Done" looks like:''' You have a summary of two contrasting expert viewpoints and a clear understanding of the key trade-off in your decision.

----

== 🧭 Why this matters (Strategists start here) ==

This exercise builds the foundational skill behind all multi-agent AI workflows: '''defining specialized roles and comparing their outputs'''. At the intermediate level, you'll split these roles across separate AI sessions with different contexts. At the advanced level, you'll design entire agent architectures. But it all starts here — training yourself to think in terms of roles, perspectives, and structured disagreement rather than asking AI once and accepting the first answer.

----

== Reflection ==
* Did one perspective feel stronger than the other? Why — was it genuinely better argued, or did it just align with what you already believed?
* What did the facilitator synthesis surface that you hadn't considered?
* Would you use this dual-role technique for real decisions going forward? What types of decisions benefit most?
* 💬 ''Run this exercise with a colleague in the room. Have them choose different expert roles than you did for the same problem — the role selection itself reveals different priorities.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[Exercises/Agent Collaboration/Ac Intermediate 01|AC-Intermediate-01]] — where you'll split these roles across separate AI sessions and learn to manage handoffs between them.

Back to [[Agent Collaboration|Agent Collaboration]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Agent Collaboration Exercises]]

Workflow Automation

2026-03-16T16:23:22Z

Admin: Fix 2 internal link(s)

''Building repeatable, AI-assisted processes that save time and reduce manual effort. Where AI fluency turns into tangible productivity.''

Building repeatable, AI-assisted processes that save time and reduce manual effort. This pillar focuses on identifying automation opportunities, designing workflows, and integrating AI into existing processes.

'''Community average score: 65%''' — solid middle ground. Most users are past basic and approaching intermediate. The gap is usually between "I use AI when I think of it" and "I've designed how AI fits into my recurring work."

== Why this pillar matters ==

Most people use AI in one-off conversations that vanish. They ask a question, get an answer, maybe use it, and start from scratch next time. That's valuable but inefficient — like rewriting the same email from scratch every week instead of using a template.

Workflow Automation is about making your AI usage ''systematic''. Instead of ad-hoc prompts, you build reusable templates. Instead of single queries, you chain steps into pipelines. Instead of doing everything yourself, you design processes where AI handles the predictable parts and you handle the judgment calls.

For generalists, this is where AI fluency becomes tangible. You're not just "good at prompting" — you've identified the 3 tasks you do weekly that don't need your brain, and you've automated 2 of them. The time you save goes to the work that actually requires you: empathy, judgment, strategy, relationships.

== What this looks like at each level ==

=== Basic — Reusable templates ===

You're learning to capture what works. Instead of typing a fresh prompt every time you summarize meeting notes, you have a template with placeholders: paste in the notes, get a consistent summary. The core skill: separating what stays the same from what changes.

'''What it feels like:''' You have a saved prompt template for at least one recurring task. You've tested it with different inputs and refined it. You share it with a colleague and it works for them without explanation.

=== Intermediate — Prompt chains ===

You've moved from single prompts to multi-step workflows. You decompose a task into stages (research, draft, refine), give each stage a specialized AI role, and pass outputs between them. The core skill: designing information flow between steps.

'''What it feels like:''' You produce a deliverable that went through a 3-step AI pipeline. You notice where context gets lost between steps and design handoffs to preserve it. You document the chain so you can reuse it.

=== Advanced — Production workflows ===

You're designing end-to-end processes with quality gates, error handling, and documentation. You map which steps are human vs. AI, define verification checkpoints, and create blueprints that others can run. The core skill: building systems, not just using tools.

'''What it feels like:''' You've mapped a business process, identified which steps AI can handle, and built production-ready prompt templates for each. You've measured time savings. Someone else can run your workflow without your involvement.

== Common mistakes ==
* '''Automating the wrong things.''' The best automation candidates are tasks that are repetitive, predictable, and low-stakes. High-judgment, novel, or high-stakes tasks should stay human-driven (with AI support, not AI control).
* '''Skipping quality gates.''' Chaining AI steps without verification is like building a pipeline without pressure checks. Each step should have a way to catch bad output before it flows downstream.
* '''Over-engineering early.''' Start with one reusable template that saves you 10 minutes a week. That's more valuable than an elaborate multi-agent system that you never finish building.

== How this connects to other pillars ==
* '''[[Prompt Engineering Basics|Prompt Engineering Basics]]''' — every workflow step is a prompt. Better prompts mean better workflows.
* '''[[Pillars/Ethical Prompting|Ethical Prompting]]''' — automated workflows need quality gates and verification at every stage
* '''[[Agent Collaboration|Agent Collaboration]]''' — agent workflows are automation taken to the next level, with AI deciding its own next steps

== Exercises ==

| Level || Exercise || Time || What you'll build

| Basic || [[Exercises/Workflow Automation/Wa Basic 01 || The Reusable Prompt]] || 15 min || A prompt template for a recurring task
| Intermediate || [[Exercises/Workflow Automation/Wa Intermediate 01 || The Prompt Chain]] || 25 min || A multi-step AI pipeline
| Advanced || [[Exercises/Workflow Automation/Wa Advanced 01 || The Workflow Blueprint]] || 40 min || A complete AI-automated business process

[[Category:AI Fluency Playbook]]
[[Category:Five Pillars]]

Why AI Gets Things Wrong

2026-03-16T16:23:21Z

Admin: Fix 2 internal link(s)

''AI doesn't lie — it generates plausible text. Understanding why helps you catch mistakes before they matter.''

<blockquote>
'''Plain English:''' AI produces confident-sounding text that is sometimes completely wrong. This isn't a bug — it's a fundamental feature of how these systems work. Knowing why it happens is the single most important thing you can learn about working with AI.
</blockquote>

== What's actually happening ==

When AI writes something incorrect, people call it a "hallucination." The word is a bit misleading — it implies the AI is seeing things that aren't there, like a glitch. What's actually happening is simpler and more important to understand:

AI generates the most statistically likely next words based on patterns in its training data. It has no mechanism to check whether what it's producing is true. It doesn't "know" things the way you know your own phone number. It produces text that ''looks and sounds like'' correct text, because it learned from millions of examples of correct text.

This means:
* It can write a perfectly formatted citation for a paper that doesn't exist — because it's learned the ''pattern'' of what citations look like.
* It can confidently state a statistic that's completely fabricated — because it's generating a plausible number in a plausible context.
* It can describe a product feature that was never built — because the description sounds like something that ''could'' exist.

The AI isn't lying to you. Lying requires knowing the truth and choosing to say something different. AI doesn't know the truth. It's generating plausible text. That's a crucial distinction.

== When it's most likely to go wrong ==

Hallucinations aren't random. They follow predictable patterns:

'''Specific facts, numbers, and dates.''' Ask AI for a general explanation of how photosynthesis works and it'll be accurate. Ask it for the exact year a specific obscure paper was published and it might invent one. The more specific and verifiable the claim, the more you need to check it.

'''Citations and sources.''' AI is particularly bad at this. It will confidently produce author names, paper titles, journal names, and URLs that look real but don't exist. Never trust an AI-generated citation without verifying it.

'''Recent events.''' AI models have a training cutoff date. If you ask about something that happened after that date and the AI doesn't have search access, it may either say it doesn't know (good) or generate a plausible-sounding answer (dangerous).

'''Niche or specialized domains.''' AI performs best on topics that appeared frequently in its training data. Mainstream topics in English have dense coverage. Obscure or specialized topics — especially in other languages — have less, so the model has fewer patterns to draw from and is more likely to fill gaps with plausible-sounding fabrications.

'''When you push it.''' If you insist the AI answer a question it's uncertain about, or tell it "you must provide an answer," it will comply — by generating something. AI tools generally don't have a strong instinct to say "I don't know." Some are better than others, but the pressure to produce output is built into the system.

== Why it ''sounds'' so confident ==

This is the part that trips people up. When a human says something confidently, you assume they believe it and probably have some basis for it. When AI says something confidently, it means nothing — confidence is the default mode.

The model isn't more certain about accurate statements than inaccurate ones. It generates all text with the same fluent, authoritative tone because that's the pattern in its training data. Well-written text sounds confident. The model produces well-written text. Therefore, everything it produces sounds confident — including the wrong things.

This is why the philosopher Harry Frankfurt's essay "On Bullshit" is on our [[Further Reading|Further Reading]] page. Frankfurt distinguishes between lying (knowing the truth and hiding it) and bullshitting (not caring whether something is true). AI is, technically, the world's most sophisticated bullshit generator. Not because it's trying to deceive you, but because truth and falsehood aren't categories it operates in. It operates in plausibility.

== What you can do about it ==

The good news: once you understand this, you can work with it effectively. The [[Exercises/Ethical Prompting/Ep Basic 01|Fact-Check Habit]] and [[Exercises/Ethical Prompting/Ep Intermediate 01|Verification Checklist]] exercises build these skills in practice. Here's the framework:

'''Treat AI output as a first draft, not a final answer.''' This shift in mindset is the most important thing. AI gives you a starting point — fast, broad, often useful. Your job is to validate, refine, and apply judgment.

'''Cross-reference specific claims.''' If the AI states a fact, a statistic, or a date that matters to your work, verify it with a primary source. This takes 30 seconds and prevents the kind of embarrassing errors that erode trust.

'''Watch for the patterns above.''' You now know when hallucinations are most likely. Apply extra scrutiny to specific facts, citations, recent events, and niche topics.

'''Ask the AI to flag its uncertainty.''' Adding "If you're not sure about something, say so" to your prompt doesn't guarantee honesty, but it does help some models hedge appropriately rather than fabricating with confidence.

'''Use AI features that ground responses in sources.''' Tools like Perplexity, Claude with web search, or ChatGPT with browsing can cite where they found information. This doesn't eliminate errors, but it gives you something to check against.

'''Don't ask AI to be your only source.''' AI is best when it's one input among several — when you're using it alongside your own expertise, your colleagues' perspectives, and verified data. The [[Exercises/Insight Synthesis/Is Intermediate 01|Multi-Source Brief]] exercise practices exactly this.

== The bigger picture ==

Understanding hallucinations isn't just about catching mistakes. It fundamentally shapes how you think about AI's role in your work:
* It's why the [[Pillars/Ethical Prompting|Ethical Prompting & Judgment]] pillar exists — because using AI responsibly requires knowing its limitations.
* It's why human oversight matters in any AI workflow — and why the most advanced exercises in this playbook always include human checkpoints.
* It's why AI fluency is more than just "knowing how to prompt" — it's knowing when to trust, when to verify, and when to override.

== Where to go next ==
* [[Exercises/Ethical Prompting/Ep Basic 01|The Fact-Check Habit]] — practice catching AI mistakes in 15 minutes
* [[Prompt Engineering Basics|Prompt Engineering Basics]] — techniques that reduce (but never eliminate) hallucinations

[[Category:AI Fluency Playbook]]
[[Category:Core Concepts]]

What We Mean by AI Fluency

2026-03-16T16:23:21Z

Admin: Fix 11 internal link(s)

''AI fluency isn't about mastering tools or memorizing prompts. It's about learning to think alongside AI — consistently, critically, and in your own context.''

<blockquote>
'''For:''' Anyone who wants to understand what AI fluency actually is, why it matters for generalists, and how this playbook helps you build it. This is the page you send to your manager, your team, or your skeptical friend.
</blockquote>

== AI fluency is not what you think it is ==

Let's start with what AI fluency is ''not'':
* '''Not "prompt engineering."''' Prompting is one skill within a much broader capability. Knowing how to phrase a request is useful. Knowing when to trust the response, when to push back, and how to integrate AI into your actual work — that's fluency.
* '''Not "knowing how AI works technically."''' You don't need to understand transformer architectures or attention mechanisms to use AI well. (Though if you're curious, [[How AI Actually Works|How AI Actually Works]] covers the essentials, and [https://xuecodex.tech/docs XueCodex] goes deep.)
* '''Not "using ChatGPT sometimes."''' Pasting text into an AI tool and accepting what comes back is AI awareness, not fluency. It's the difference between knowing a language exists and actually being able to think in it.

Here's what we mean:

<blockquote>
'''AI fluency is the ability to work with AI effectively, ethically, and intentionally across your real work — not as a novelty, but as a core part of how you think, decide, and deliver.'''
</blockquote>

The key words are ''intentionally'' and ''real work''. An AI-fluent person doesn't just use AI when it's convenient. They know which parts of their work benefit from AI, which don't, and they can explain why. They have habits, not just tricks.

== Five dimensions of fluency ==

We've broken AI fluency into five concrete capabilities. These aren't abstract categories — they're things you ''do'':

| Pillar || What it means || What it looks like

| '''[[Insight Synthesis|| Insight Synthesis]]''' || Extracting meaning from noise || You use AI to surface patterns across 50 customer feedback entries, then apply your domain knowledge to decide what actually matters
| '''[[Workflow Automation|| Workflow Automation]]''' || Designing AI-augmented processes || You've identified the 3 tasks you do weekly that don't need your brain, and you've automated 2 of them
| '''[[Cross-Domain Reframing|| Cross-Domain Reframing]]''' || Bridging perspectives and adapting ideas || You translate a technical AI recommendation into language your finance team can evaluate and act on
| '''[[Agent Collaboration|| Agent Collaboration]]''' || Working alongside AI with clear roles || You've set up AI as a persistent collaborator that knows your work context, not a blank chatbot you re-explain things to every time
| '''[[Pillars/Ethical Prompting || Ethical Prompting & Judgment]]''' || Responsible, transparent, critical use || You can explain to a stakeholder why you trusted an AI output in one case and overrode it in another

These aren't separate skills you learn in sequence. They overlap and reinforce each other. Every real AI-fluent action involves two or three of these at once. When you use AI to synthesize research (Insight Synthesis), reframe it for a different audience (Cross-Domain Reframing), and verify its claims before publishing (Ethical Prompting) — that's fluency in action.

=== How this connects to other frameworks ===

We didn't invent the idea of AI fluency. The 4D framework — Delegation, Description, Discernment, Diligence — developed by Prof. Joseph Feller and Prof. Rick Dakan and taught in [https://anthropic.skilljar.com/ai-fluency-framework-foundations Anthropic's free AI Fluency course] covers similar territory from an academic foundations angle. UNESCO's AI competency frameworks and HR competency models that now include AI fluency alongside data literacy echo the same patterns.

The convergence is telling: across different frameworks, effective AI use requires both practical capability ''and'' judgment. Our five pillars are one way to organize that — designed specifically for generalists who need to practice, not just understand.

A rough mapping:
* '''Delegation''' (deciding what to hand to AI) ↔ Agent Collaboration & Workflow Automation
* '''Description''' (communicating clearly with AI) ↔ Insight Synthesis & Cross-Domain Reframing
* '''Discernment & Diligence''' (evaluating and verifying) ↔ Ethical Prompting & Judgment

If you want the academic foundations, take Anthropic's free course — it's excellent. This playbook picks up where courses leave off: it's where you ''practice'' fluency in your actual work.

== Why this matters for generalists specifically ==

AI fluency courses and frameworks are everywhere. Most of them target developers, data scientists, or "everyone" — which in practice means they're either too technical or too generic. Here's why generalists need something different.

=== You can't opt out ===

Modern competency models now place AI fluency alongside business acumen and data literacy as a core capability. Specialists can get by without it for a while because their deep domain expertise carries them. Generalists don't have that luxury. Your value comes from working across multiple domains, and AI is now part of every one of them. Marketing, operations, strategy, project management, communications. It's already there, whether you invited it or not.

=== The real shift isn't speed ===

Yes, AI fluency lets you handle routine work faster. But the bigger change is what you do with the reclaimed time and attention. The AI-fluent generalist doesn't just work faster; they work on different things. They focus on empathy, ethical reasoning, contextual judgment, relationship building — the capabilities AI can't replace.

That's the shift — not doing the same work faster, but doing different work entirely.

=== Someone has to maintain standards ===

Generalists oversee processes across teams. When AI is involved in those processes — and increasingly it is — someone needs to ensure quality and accountability. Three capabilities matter:
* '''Decision ownership:''' Knowing when AI is advisory and when the human is accountable
* '''Interpretation:''' Assessing whether AI output is relevant and accurate ''in this specific context''
* '''Transparency:''' Being able to explain AI-assisted decisions to people who didn't see the process

These map directly to our [[Pillars/Ethical Prompting|Ethical Prompting & Judgment]] pillar, and they're the reason that pillar exists even though it's where people score highest on the quiz. Confidence without rigor is the most dangerous pattern in AI use.

=== You spot opportunities others miss ===

Generalists work across departments. AI fluency helps you spot automation opportunities, collaboration patterns, and synthesis needs that specialists in one domain might not see. A marketing person doesn't notice the overlap between their weekly competitor report and the sales team's pipeline analysis. A generalist who works with both does — and can connect them with AI.

This is [[Cross-Domain Reframing|Cross-Domain Reframing]] in action, and it's one of the most valuable things a generalist brings to any organization.

=== The gap is widening ===

There's a growing divide between passive users ("I paste into ChatGPT and use whatever comes back") and active shapers ("I've designed how AI fits into my team's workflow"). AI-fluent generalists become the people who lead adoption, not just follow it. They're the ones who say "here's how we should use this" rather than "I guess we should try this." That's the difference between using a tool and being fluent in a capability.

== How this playbook helps you build it ==

This isn't a course. There's no start-to-finish curriculum, no final exam, no certificate. It's a handbook — designed to be picked up when you need it, started wherever makes sense, and revisited as you grow.

Here's how the pieces fit together:

'''Self-assessment → targeted practice → reflection'''
# '''Start with the quiz.''' The [https://aiskillsquiz.com AI Skills Quiz] takes a few minutes and maps your current strengths across all five pillars. It tells you where you have the most room to grow — and for most people, that's [[Agent Collaboration|Agent Collaboration]] (community average: 51%).
# '''Read the pillar that matters most.''' Each pillar page is a complete guide: what the capability looks like in real work, common myths, how the levels feel from the inside, and real stories from people like you. You don't have to start at basic — start where you are.
# '''Do an exercise.''' Every exercise has three entry points based on how you learn:
** 🔧 '''Jump in''' — for people who learn by doing (42% of quiz takers)
** 📋 '''Plan first''' — for people who want structure before action (25%)
** 🧭 '''Why this matters''' — for people who need to understand the strategic context (23%)
# '''Reflect and connect.''' Exercises include reflection questions that help you see how this connects to your real work and other pillars. This is where learning turns into lasting capability.
# '''Read further.''' Curated resources — all human-vetted, no AI slop — deepen specific topics when you're ready.

The concept pages ([[How AI Actually Works|How AI Actually Works]], [[Why AI Gets Things Wrong|Why AI Gets Things Wrong]], [[Prompt Engineering Basics|Prompt Engineering Basics]]) give you the foundational knowledge that makes everything else click.

== Where to start ==

Take the [https://aiskillsquiz.com AI Skills Quiz] for a personalized recommendation, or start with your lowest-scoring pillar — that's where you'll see the most growth. If you want to understand the foundations first, read the concept pages starting with [[How AI Actually Works|How AI Actually Works]].

----

''This playbook is part of [https://generalistworld.com Generalist World]. It's open, evolving, and built on the belief that AI fluency is for everyone — not just engineers.''

[[Category:AI Fluency Playbook]]
[[Category:Core Concepts]]

Tools and Platforms

2026-03-16T16:23:20Z

Admin: Fix 2 internal link(s)

''Recommended AI tools and platforms for generalists, with honest assessments of strengths and limitations.''

All exercises in this playbook work with any AI chat tool. You don't need a specific platform — use whatever you're comfortable with. This page lists options by category for reference.

== AI Chat Tools ==

These are the tools you'll use for most exercises. Any one of them is sufficient.

| Tool || Best For || Notes

| '''ChatGPT''' (OpenAI) || General-purpose exercises, wide model selection || Free tier available. GPT-4o recommended for more complex exercises.
| '''Claude''' (Anthropic) || Long-form analysis, nuanced reasoning || Strong at synthesis and following complex instructions.
| '''Gemini''' (Google) || Integration with Google Workspace, multimodal tasks || Good for exercises involving documents or data in Google ecosystem.
| '''Copilot''' (Microsoft) || Integration with Microsoft 365 || Useful if your workflow lives in Outlook, Word, and Teams.

== Agent Frameworks ==

Referenced in the advanced [[Agent Collaboration|Agent Collaboration]] exercises. '''Not required for any exercise''' — the playbook teaches agent thinking through manual prompting first.

| Framework || What It Does || When to Explore

| '''CrewAI''' || Python framework for orchestrating multiple AI agents with defined roles || After completing [[Exercises/Agent Collaboration/Ac Advanced 01 || AC-Advanced-01]] — when you want to automate the handoffs you did manually
| '''AutoGen''' (Microsoft) || Framework for building multi-agent conversations || When you want agents that can talk to each other without manual copy-pasting
| '''LangGraph''' (LangChain) || Framework for building stateful agent workflows with graph-based logic || When you need complex conditional logic in your agent pipelines

== Automation Tools ==

Relevant to the [[Workflow Automation|Workflow Automation]] pillar, especially at the advanced level.

| Tool || What It Does || When to Explore

| '''n8n''' || Open-source workflow automation with AI integration nodes || After completing [[Exercises/Workflow Automation/Wa Advanced 01 || WA-Advanced-01]] — when you want to automate your prompt chains
| '''Make''' (formerly Integromat) || Visual workflow builder with AI steps || Good for non-technical users who want to automate without code
| '''Zapier''' || Simple automation connectors between apps, with AI actions || Best for straightforward automations: trigger → AI step → action

== Productivity Integrations ==

Tools that embed AI into existing workflows.

| Tool || What It Does

| '''Notion AI''' || AI writing and analysis built into Notion workspace
| '''Obsidian + Smart Connections''' || AI-powered linking and search within an Obsidian vault
| '''Raycast AI''' || Quick AI access from any application on macOS
| '''Google Workspace AI''' || AI features embedded in Docs, Sheets, Gmail

[[Category:AI Fluency Playbook]]
[[Category:Resources]]

Tokenization & Context Windows

2026-03-16T16:23:19Z

Admin: Fix 1 internal link(s)

''Why AI can only 'remember' so much, why long conversations go off the rails, and what you can do about it.''

<blockquote>
'''Plain English:''' AI doesn't read words — it reads "tokens" (word chunks). Every AI tool has a limit on how many tokens it can handle at once. This is the context window, and it's the single biggest constraint on what AI can do for you.
</blockquote>

== What tokens actually are ==

When you type a message to AI, it doesn't see words the way you do. It breaks your text into smaller pieces called '''tokens'''. Sometimes a token is a whole word. Sometimes it's a piece of a word. Sometimes it's just punctuation.

For example:
* "Hello" → 1 token
* "Tokenization" → might become "Token" + "ization" → 2 tokens
* "I'm building a workflow" → roughly 5 tokens

A rough rule of thumb: '''1 token ≈ ¾ of a word''' in English. So 1,000 words is roughly 1,300 tokens. A 10-page document is roughly 4,000–5,000 tokens.

Why does this matter to you? Because every AI tool charges by tokens and limits by tokens. When you hit a message length limit, get a "conversation too long" error, or notice AI "forgetting" things you said earlier — that's all about tokens.

== The context window: AI's working memory ==

The '''context window''' is the total number of tokens the AI can process at once. Think of it as a whiteboard: everything in your conversation — your messages, the AI's responses, any documents you've uploaded, the system prompt — all has to fit on this whiteboard. When it's full, things start falling off the other end.

Current context window sizes (as of early 2026):

| Model || Context window || Roughly equivalent to

| Claude (Anthropic) || 200K tokens || ~150,000 words — a full novel
| GPT-4o (OpenAI) || 128K tokens || ~96,000 words
| Gemini 1.5 Pro (Google) || 1M+ tokens || ~750,000 words — multiple books

These numbers sound enormous, but they fill up faster than you'd think. A long conversation with back-and-forth responses, a few uploaded documents, and a detailed system prompt can eat through 200K tokens in a working session.

== Why this matters for your work ==

'''Long conversations degrade.''' If you've noticed AI giving worse answers later in a conversation than at the beginning, this is why. As the context window fills up, the model has more text to process and older information gets less "attention." Starting a fresh conversation for a new topic isn't a sign of failure — it's good practice.

'''Uploaded documents have limits.''' When you upload a PDF or paste a long document, it consumes context window space. A 50-page report might use 20,000+ tokens, leaving less room for your actual questions and the AI's responses. If you're working with long documents, consider summarizing or extracting the relevant sections first.

'''"AI forgot what I said" is usually a context issue.''' AI doesn't have memory between conversations (unless you're using features like Claude's Projects or custom GPTs that provide persistent context). Even within a conversation, if you're 30 messages in, the AI may lose track of something you said at the beginning because it's being pushed out of the active window.

'''This is why the [[Exercises/Agent Collaboration/Ac Intermediate 01|Handoff Protocol]] exercise matters.''' When you learn to structure handoffs between AI sessions — summarizing context, carrying forward the essential information — you're working around context window limits intelligently.

== Practical tips ==

'''Start fresh for new topics.''' Don't keep one mega-conversation running for everything. A new topic deserves a new conversation with focused context.

'''Front-load the important stuff.''' Put your most critical instructions, context, and constraints at the beginning of your prompt. Information at the start and end of the context window gets more "attention" than information buried in the middle.

'''Summarize before you continue.''' If a conversation is getting long and you want to keep going, ask the AI to summarize the key decisions and context so far, then start a new conversation with that summary.

'''Use persistent context features.''' Claude Projects, custom GPTs, and system prompts let you set context that persists across messages without eating into your per-message token budget. The [[Exercises/Agent Collaboration/Ac Intermediate 01|Handoff Protocol]] exercise teaches you to design these.

'''Be selective with document uploads.''' Instead of uploading a 100-page document and asking a question, extract the 5 relevant pages. You'll get better answers and use less of your context budget.

== Where to go next ==
* [[Prompt Engineering Basics|Prompt Engineering Basics]] — how to make the most of limited context
* [[Exercises/Agent Collaboration/Ac Intermediate 01|The Handoff Protocol]] — practice managing context across AI sessions

[[Category:AI Fluency Playbook]]
[[Category:Core Concepts]]

The Workflow Blueprint

2026-03-16T16:23:18Z

Admin: Fix 1 internal link(s)

''Design, document, and test a complete AI-automated workflow for a real business process. 40 minutes.''

<blockquote>
'''One-liner:''' Design, document, and test a complete AI-automated workflow for a real business process — from trigger to output, with error handling and quality gates.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a real business process that currently takes you 30+ minutes and involves multiple steps. Examples: weekly reporting, content production, customer onboarding documentation, project status updates, invoice processing.

'''Phase 1 — Map the current process.''' Send this:

<blockquote>
I'm going to automate this business process: '''[describe the process]'''.

Help me map the current manual workflow:
1. What triggers the process? (time-based, event-based, request-based)
2. What are the sequential steps from trigger to final output?
3. What inputs does each step require?
4. What decisions are made at each step? (if/then logic)
5. Where are the bottlenecks or error-prone points?
6. What's the final deliverable and who receives it?

Present this as a numbered workflow with decision points marked.
</blockquote>

'''Phase 2 — Design the AI workflow.''' Send:

<blockquote>
Now redesign this as an AI-automated workflow. For each step, specify:

| Step || Human or AI? || If AI: what prompt template? || If Human: what decision? || Input || Output || Quality gate

Rules:
- Some steps should remain human (judgment calls, approvals, sensitive decisions)
- Every AI step needs a quality gate — how do you know the output is good enough to proceed?
- Include error handling — what happens when an AI step produces bad output?
- Include a feedback mechanism — how does the workflow improve over time?
</blockquote>

'''Phase 3 — Write the prompt templates.''' For each AI step in the workflow:

<blockquote>
Write the production-ready prompt template for Step '''[N]''': '''[step name]'''

The template should include:
- Role definition for the AI
- Clear input specification with [PLACEHOLDERS]
- Exact output format requirements
- Quality criteria the output must meet
- An example of good output vs. bad output

This prompt should work reliably every time with different inputs. It should be usable by someone who didn't design the workflow.
</blockquote>

'''Phase 4 — Run the workflow end-to-end.''' Execute the full pipeline with real data. Track:
* Time per step (manual vs. AI-assisted)
* Quality gate pass/fail rates
* Where you had to intervene or override
* Total time saved vs. the manual process

'''Phase 5 — Document the blueprint.''' Create a 1-page workflow document:

<blockquote>
Write a "Workflow Blueprint" for this process that includes:
1. '''Trigger:''' What starts the workflow
2. '''Flow diagram:''' Step-by-step with decision points (use text-based flowchart)
3. '''Prompt templates:''' Reference to each template (step number and name)
4. '''Quality gates:''' What to check at each stage
5. '''Error handling:''' What to do when something fails
6. '''Maintenance:''' How to update the workflow as requirements change
7. '''Metrics:''' How to measure whether the workflow is working well
</blockquote>

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a process''' — Pick something that takes 30+ minutes, involves multiple steps, and happens regularly. The more manual the current process, the bigger the payoff.
# '''Map the current workflow''' — Document every step, decision point, and handoff. You can't automate what you don't understand.
# '''Design the hybrid workflow''' — Decide what AI handles vs. what stays human. Add quality gates and error handling. Not everything should be automated.
# '''Build the prompt templates''' — Write production-grade prompts for each AI step. These should be reusable by anyone, not just you.
# '''Test end-to-end''' — Run the full workflow with real data. Measure time, quality, and failure points.
# '''Document the blueprint''' — Create a shareable document that anyone could use to run this workflow.

'''"Done" looks like:''' A complete, tested workflow blueprint with prompt templates, quality gates, and measured time savings. Something you could hand to a colleague and they could execute without additional explanation.

----

== 🧭 Why this matters (Strategists start here) ==

In [[Exercises/Workflow Automation/Wa Intermediate 01|WA-Intermediate-01]], you built a 3-step prompt chain. Here, you're building a '''production-grade workflow''' — the kind of thing that saves hours per week and can be delegated. The key differences from an intermediate prompt chain: quality gates (not just chaining outputs blindly), error handling (what happens when AI fails), and documentation (others can run it without you). This is directly transferable to tools like n8n, Make, or Zapier with AI steps. The blueprint format is also the deliverable that organizations pay consultants to produce.

----

== Reflection ==
* How much time did the automated workflow save compared to the manual process? Is the saving worth the design effort?
* Which quality gates caught real problems? Which were unnecessary overhead?
* Where did AI fail and require human override? Was that predictable from the design phase, or did it only emerge during testing?
* 💬 ''Walk a colleague through your workflow blueprint and ask them to find the step most likely to fail. Fresh eyes spot single points of failure you've normalized.'' (Social Learners)

== ⬆️ Level up ==

You've reached the advanced level for Workflow Automation. From here, consider:
* Implementing this workflow in an automation tool (n8n, Make, Zapier) for true hands-free execution
* Combining this with [[Exercises/Agent Collaboration/Ac Advanced 01|AC-Advanced-01]] to add multi-agent architecture to your workflow steps
* Measuring workflow performance over 4 weeks and iterating based on failure data

Back to [[Workflow Automation|Workflow Automation]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Workflow Automation Exercises]]

The Stolen Technique

2026-03-16T16:23:17Z

Admin: Fix 1 internal link(s)

''Take an AI technique from a completely different field and apply it to your own work. 15 minutes.''

<blockquote>
'''One-liner:''' Take an AI technique from a completely different field and apply it to your own work — discovering that the best prompting ideas are often borrowed.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a field that is '''not''' your own. If you work in marketing, pick engineering. If you're a designer, pick finance. If you're a developer, pick journalism. The more unfamiliar, the better.

'''Step 1 — Discover a technique.''' Send this prompt:

<blockquote>
How do professionals in '''[unfamiliar field]''' use AI in their daily work? Give me 5 specific, concrete techniques — not general concepts. For each technique, describe: what they prompt the AI to do, what input they provide, and what output they get. Focus on techniques that are unique to this field.
</blockquote>

'''Step 2 — Steal the best one.''' Pick the technique that seems most interesting or most different from how you currently use AI. Then send:

<blockquote>
I work in '''[your field]'''. Take the technique you described as #[number] — '''[briefly describe it]''' — and help me adapt it for my work. Specifically:
1. What would the equivalent input look like in my field?
2. How would I modify the prompt to fit my context?
3. What output would I expect?
4. Write me a ready-to-use prompt that applies this borrowed technique to '''[a specific task you do]'''.
</blockquote>

'''Step 3 — Test it.''' Copy the adapted prompt. Use it on a real task. Compare the result to how you'd normally approach it.

'''Example — a marketer borrowing from investigative journalism:'''

The technique: Journalists use AI to cross-reference claims across multiple sources and flag inconsistencies.

The adaptation: A marketer uses the same technique to cross-reference their product claims against competitor claims and customer reviews, flagging gaps between promise and reality.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Pick an unfamiliar field''' — Choose something genuinely outside your expertise. The discomfort is the point — that's where non-obvious ideas live.
# '''Research AI techniques in that field''' — Use AI to discover how professionals in that domain use AI tools. Look for specific techniques, not generalities.
# '''Identify a transferable technique''' — Pick one that solves a problem similar to something in your work, even though it looks completely different on the surface.
# '''Adapt with AI's help''' — Ask the AI to bridge the gap between the source domain and your domain. Get a ready-to-use prompt.
# '''Test the borrowed technique''' — Apply it to a real task and evaluate whether it gives you a different (and possibly better) result than your usual approach.

'''"Done" looks like:''' You have a working prompt borrowed from another field that gives you a new angle on a familiar task.

----

== 🧭 Why this matters (Strategists start here) ==

Most people prompt AI using patterns from their own field — but the most powerful AI techniques are often domain-agnostic. Researchers structure AI analysis differently than marketers, engineers test AI outputs differently than writers, and each field has developed prompting patterns the others rarely see. Cross-domain reframing is how you break out of local optima in your AI usage. At the intermediate level, you'll systematically adapt entire prompt strategies across domains; this exercise builds the muscle of looking outside your field for AI inspiration.

----

== Reflection ==
* Did the borrowed technique produce a noticeably different result than your usual approach? Better, worse, or just different?
* What made the technique transferable? Was it the structure, the question type, or the underlying problem it solves?
* Which other field would you explore next for AI techniques? What made you choose it?
* 💬 ''Ask a colleague from a different department how they use AI. You'll likely discover a technique you've never considered — that's cross-domain reframing in action.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[Exercises/Cross Domain Reframing/Cdr Intermediate 01|CDR-Intermediate-01]] — where you'll systematically adapt an entire prompting strategy from an unfamiliar domain.

Back to [[Cross-Domain Reframing|Cross-Domain Reframing]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Cross-Domain Reframing Exercises]]

The Signal in the Noise

2026-03-16T16:23:16Z

Admin: Fix 1 internal link(s)

''Turn a messy AI brainstorm into structured, actionable insight. 15 minutes.''

<blockquote>
'''One-liner:''' Turn a messy AI brainstorm into a structured, actionable insight — learning to extract what matters and discard what doesn't.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a topic you're genuinely curious about or working on. It could be a business challenge, a learning goal, or a decision you need to make.

'''Step 1 — Generate the mess.''' Send this prompt to any AI:

<blockquote>
Brainstorm 15-20 ideas about '''[your topic]'''. Don't filter or organize — just generate as many ideas as possible, even contradictory or half-formed ones. Number each idea.
</blockquote>

'''Step 2 — Extract the signal.''' Now send this follow-up:

<blockquote>
Look at the brainstorm you just generated. Identify:
1. '''The top 3 ideas''' that are most actionable within the next week
2. '''The 1 idea''' that's most surprising or non-obvious
3. '''The 2 ideas''' that contradict each other — and what the tension between them reveals
4. '''The pattern''' — what theme or assumption connects most of these ideas?

For each, explain your reasoning in one sentence.
</blockquote>

'''Step 3 — Challenge the synthesis.''' Send this:

<blockquote>
Now tell me what's missing from this brainstorm. What obvious angle or perspective did you fail to include? Add 3 ideas that fill that gap.
</blockquote>

Read the final output. You started with noise; you now have structured insight. The skill here isn't prompting — it's knowing what questions to ask ''after'' the AI generates raw material.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a topic''' — Something you care about. The exercise works best with real problems, not hypotheticals.
# '''Generate raw material''' — Ask AI for a large, unfiltered brainstorm (15-20 ideas). The messier the better — that's the point.
# '''Apply a synthesis framework''' — Use the structured follow-up prompt to force the AI to categorize, rank, and find patterns in its own output.
# '''Identify gaps''' — Ask the AI what it missed, then evaluate whether the gap-filling ideas actually change your understanding.
# '''Capture your insight''' — Write a single sentence summarizing what you learned that you didn't know before.

'''"Done" looks like:''' You have 3 actionable ideas, 1 non-obvious insight, a clear tension to think about, and a unifying pattern — extracted from a wall of brainstorm text.

----

== 🧭 Why this matters (Strategists start here) ==

AI is excellent at generating volume but mediocre at distinguishing signal from noise — that's still a human skill. This exercise builds your ability to use AI as a '''thinking amplifier''' rather than an answer machine. The synthesis framework (rank, surprise, contradict, pattern) is reusable: apply it to research outputs, meeting notes, customer feedback analysis, or any situation where you need to extract meaning from quantity. At the intermediate level, you'll synthesize across ''multiple'' AI outputs; this exercise builds the foundation.

----

== Reflection ==
* Did the AI's ranking match your instinct? Where did you disagree, and what does that tell you about the AI's priorities vs. yours?
* Was the "gap" the AI identified actually a meaningful blind spot, or was it filler?
* Would you use this brainstorm-then-synthesize pattern again? For what kinds of problems does it work best?
* 💬 ''Run the same brainstorm prompt with a colleague present. Compare which ideas you each gravitate toward — the difference reveals your respective assumptions.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[Exercises/Insight Synthesis/Is Intermediate 01|IS-Intermediate-01]] — where you'll synthesize across multiple AI sessions to build a more complete picture.

Back to [[Insight Synthesis|Insight Synthesis]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Insight Synthesis Exercises]]

The Reusable Prompt

2026-03-16T16:23:15Z

Admin: Fix 1 internal link(s)

''Turn a repeatable task into a reusable AI prompt template. Your first step toward automation. 15 minutes.''

<blockquote>
'''One-liner:''' Turn a task you do repeatedly into a reusable AI prompt template that works every time — your first step toward automation.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Think of something you do at least once a week that involves writing, analyzing, or summarizing. Examples: writing a status update, summarizing meeting notes, drafting an email to a client, reviewing a document.

'''Step 1 — Do it once with AI.''' Open any AI chat and do the task the way you normally would — just ask the AI to help. Don't overthink the prompt. Just get the job done.

'''Step 2 — Reverse-engineer your prompt.''' Now send this:

<blockquote>
Look at the prompt I just gave you and the output you produced. Help me turn this into a '''reusable template''' that I can use every time I need to do this task. The template should have:
1. '''Clear placeholders''' — marked with [BRACKETS] for the parts that change each time
2. '''Fixed instructions''' — the parts that stay the same every time
3. '''Output format specification''' — exactly what the result should look like (length, structure, tone)

Write the template so someone else on my team could use it without any additional explanation.
</blockquote>

'''Step 3 — Test it.''' Copy the template. Start a new chat. Paste the template and fill in the placeholders with a different example of the same task. Does the output match the quality of your original? If not, adjust the template.

'''Here's a concrete example:'''

''Original task:'' "Summarize this meeting for my team"

''Reusable template:''
<blockquote>
Summarize the following meeting notes for a team update.

'''Meeting notes:''' [PASTE NOTES HERE]

'''Output requirements:'''
- Start with a 1-sentence summary of the main decision or outcome
- List action items with owner names in bold
- Flag any unresolved questions
- Keep the total summary under 150 words
- Tone: professional but informal
</blockquote>

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Identify a repeatable task''' — Pick something you do weekly that involves text: writing, summarizing, analyzing, or formatting. The more repetitive, the better.
# '''Do it once with AI''' — Complete the task normally. Don't try to be clever — just get a result you're happy with.
# '''Extract the template''' — Ask the AI to help you identify what's fixed (instructions, format, tone) vs. what changes (the input data). Build a reusable template with clear placeholders.
# '''Test with a new example''' — Use the template on a fresh instance of the same task. Compare quality to the original.
# '''Refine if needed''' — If the template didn't produce equally good output, identify what was missing and add it.

'''"Done" looks like:''' You have a saved prompt template with clear placeholders that consistently produces good output for your repeatable task.

----

== 🧭 Why this matters (Strategists start here) ==

Most people use AI in one-off conversations that disappear. This exercise introduces the shift from '''ad-hoc prompting to systematic workflows''' — the foundation of all AI automation. A reusable template is the simplest form of an AI workflow: defined input, consistent process, predictable output. At the intermediate level, you'll chain multiple templates together into multi-step workflows. Every automated AI process in production started as someone's reusable prompt.

----

== Reflection ==
* What did you have to add to the template that wasn't obvious from the original prompt?
* Did the template produce consistent quality with different inputs, or did you need to tweak it? What was missing?
* How much time will this template save you per week? Is it enough to justify the setup effort?
* 💬 ''Send your template to a colleague who does the same task. Can they use it without any explanation? Their confusion points reveal where the template needs more specificity.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[Exercises/Workflow Automation/Wa Intermediate 01|WA-Intermediate-01]] — where you'll chain multiple prompt templates into a multi-step workflow.

Back to [[Workflow Automation|Workflow Automation]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Workflow Automation Exercises]]

The Research Pipeline

2026-03-16T16:23:14Z

Admin: Fix 1 internal link(s)

''Build a complete research synthesis pipeline with evidence grading and contradiction analysis. 40 minutes.''

<blockquote>
'''One-liner:''' Build a complete research synthesis pipeline — from question to evidence-graded conclusions — using structured AI queries and your own critical judgment.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a question you genuinely need answered for your work. Not a trivia question — something where the answer shapes a real decision.

'''Phase 1 — Define the research question.''' Send:

<blockquote>
I need to research this question: '''[your question]'''

Help me refine it into a research-ready question by:
1. Breaking it into 3-4 sub-questions that, if answered, would fully address the main question
2. For each sub-question, identifying what type of evidence would count as a strong answer (data, expert consensus, case studies, logical argument, etc.)
3. Flagging any assumptions embedded in the main question that I should test
</blockquote>

'''Phase 2 — Structured evidence gathering.''' For each sub-question, run a separate AI query:

<blockquote>
Research sub-question: '''[sub-question]'''

For this query, I want structured evidence:
- '''Strong evidence:''' Claims supported by widely documented data, peer-reviewed research, or established expert consensus
- '''Moderate evidence:''' Claims supported by credible case studies, industry reports, or respected analysis
- '''Weak evidence:''' Claims based on anecdotes, single examples, logical inference without data, or common assertions that may not hold up

Classify every claim you make. If you're not sure about the evidence quality, say so. I'd rather have honest uncertainty than false confidence.
</blockquote>

'''Phase 3 — Contradiction analysis.''' After running all sub-queries, send this to a fresh session:

<blockquote>
Here are the findings from my research on '''[main question]''':

'''Sub-question 1 findings:''' [paste summary]
'''Sub-question 2 findings:''' [paste summary]
'''Sub-question 3 findings:''' [paste summary]

Analyze the contradictions:
1. Where do the findings from different sub-questions conflict?
2. Which conflicts can be resolved by looking at the evidence quality?
3. Which conflicts are genuine unresolved tensions?
4. What additional evidence would resolve the remaining tensions?
</blockquote>

'''Phase 4 — Your synthesis.''' Write a 500-word research brief yourself (not AI-generated) that answers your original question. Structure it as:
# '''Bottom line:''' Your answer in 1-2 sentences
# '''Key evidence:''' The 3 strongest pieces of evidence supporting your answer, with evidence grades
# '''Key uncertainty:''' What you're least confident about and why
# '''What would change your mind:''' 1-2 pieces of evidence that, if found, would reverse your conclusion

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Formulate a research question''' — Choose something decision-relevant. Use AI to decompose it into sub-questions with defined evidence standards.
# '''Gather evidence by sub-question''' — Run separate queries for each sub-question, requiring the AI to grade its own evidence quality (strong/moderate/weak).
# '''Analyze contradictions''' — Feed all findings into a fresh session and ask for conflict analysis. Identify which conflicts are real vs. caused by weak evidence.
# '''Write your own synthesis''' — Produce a 500-word brief that answers the question, cites evidence with quality grades, and states what would change your mind.
# '''Assess the pipeline''' — Evaluate whether this process produced a meaningfully better answer than a single AI query would have.

'''"Done" looks like:''' A research brief that clearly distinguishes strong from weak evidence, acknowledges uncertainty, and provides a decision-ready answer with stated confidence.

----

== 🧭 Why this matters (Strategists start here) ==

This exercise combines the skills from [[Exercises/Insight Synthesis/Is Basic 01|IS-Basic-01]] (extracting signal from noise) and [[Exercises/Insight Synthesis/Is Intermediate 01|IS-Intermediate-01]] (triangulating across perspectives) into a '''complete research methodology'''. The evidence grading system prevents the common failure mode of treating all AI output as equally reliable. The contradiction analysis surfaces genuinely open questions rather than papering over them. This pipeline is directly applicable to due diligence, competitive intelligence, policy analysis, and any context where the cost of being wrong is high and the question is too complex for a single query.

----

== Reflection ==
* Did the evidence grading change which findings you trusted? Were you surprised by what was classified as "weak"?
* How did the contradiction analysis change your initial view?
* Was the 500-word synthesis harder or easier than expected? What was the hardest part — compression, confidence, or acknowledging uncertainty?
* 💬 ''Teach this pipeline to a colleague and have them run it on a different question. Compare how you each handle the "what would change your mind" step — that reveals different attitudes toward uncertainty.'' (Social Learners)

== ⬆️ Level up ==

You've reached the advanced level for Insight Synthesis. From here, consider:
* Using this pipeline for a real decision and tracking whether your evidence-graded conclusion held up
* Combining this with [[Exercises/Agent Collaboration/Ac Advanced 01|AC-Advanced-01]] to delegate different research phases to different agent roles
* Teaching this method to a colleague and seeing how they adapt it

Back to [[Insight Synthesis|Insight Synthesis]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Insight Synthesis Exercises]]

The Prompt Chain

2026-03-16T16:23:13Z

Admin: Fix 1 internal link(s)

''Build a multi-step AI workflow where each step's output feeds into the next. 25 minutes.''

<blockquote>
'''One-liner:''' Build a multi-step AI workflow where each step's output feeds into the next — turning a complex task into a repeatable pipeline.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a task that has at least 3 distinct phases. Examples: writing a blog post (research, outline, draft, edit), analyzing a dataset (clean, analyze, summarize, recommend), or preparing a presentation (topic research, slide structure, talking points, Q&A prep).

'''Build a 3-step chain.''' Each step is a separate prompt. The output of each step becomes the input of the next.

'''Step 1 — Research/Gather:'''
<blockquote>
You are a research assistant. Your job is to gather the raw material for '''[your task]'''.

Topic/context: '''[describe what you're working on]'''

Produce a structured collection of: key facts, relevant examples, important considerations, and any constraints. Organize by theme. Do not draft anything — just collect the ingredients.
</blockquote>

Copy the output. Start a new prompt (or clearly reset context).

'''Step 2 — Structure/Draft:'''
<blockquote>
You are a content architect. Your job is to turn raw research into a structured draft.

Here is the research material: '''[paste Step 1 output]'''

The final deliverable is: '''[describe what you need — a blog post, a report, a strategy doc, etc.]'''

Create a structured draft. Include clear sections, key arguments in order, and placeholders for any examples or data points from the research. Focus on logical flow and completeness.
</blockquote>

Copy the output. Start a new prompt.

'''Step 3 — Polish/Critique:'''
<blockquote>
You are a senior editor. Your job is to make this draft publication-ready.

Here is the draft: '''[paste Step 2 output]'''

The audience is: '''[describe who will read this]'''

Do three things:
1. Improve clarity — simplify any convoluted sentences, cut unnecessary words
2. Strengthen weak points — flag any claim that needs better support and add it
3. Check consistency — ensure tone, terminology, and formatting are uniform throughout

Produce the final version with an editor's note listing your key changes.
</blockquote>

'''Now document the chain.''' Write down the 3 prompts as a reusable template (with [PLACEHOLDERS] for the parts that change). You've just built a prompt pipeline.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a multi-phase task''' — Something that naturally has distinct stages (research → create → refine). The more phases, the more the chain helps.
# '''Design the chain''' — Write 3 prompts, each with a clear role, input expectation, and output format. The key constraint: each step's output must contain everything the next step needs.
# '''Run the chain''' — Execute each step sequentially, passing the output forward. Use fresh contexts between steps to prevent bleed-through.
# '''Evaluate information flow''' — Notice where context was lost between steps. What did Step 3 need that Step 2 didn't preserve?
# '''Document as a template''' — Save the chain with placeholders so you can reuse it for the same type of task.

'''"Done" looks like:''' A completed deliverable that went through a 3-step pipeline, plus a documented prompt chain template with placeholders for reuse.

----

== 🧭 Why this matters (Strategists start here) ==

In [[Exercises/Workflow Automation/Wa Basic 01|WA-Basic-01]], you built a single reusable prompt. Here, you're learning to '''chain prompts into a workflow''' — the building block of all production AI automation. Every AI-powered pipeline (content generation, data analysis, document processing) is fundamentally a prompt chain with handoffs. The skill you're building — decomposing a task into stages, defining clear inputs and outputs, managing context between steps — is the same skill used in tools like n8n, Zapier AI, or custom LLM pipelines. Manual chaining teaches you what to automate and where the bottlenecks live.

----

== Reflection ==
* Where did context get lost between steps? What information did a later step need that an earlier step didn't pass along?
* Did the 3-step chain produce better output than a single "do everything" prompt? Where specifically was the improvement?
* Which step in the chain was the weakest link? How would you redesign it?
* 💬 ''Have a colleague run your documented chain on a different task of the same type. Their experience reveals whether your chain is truly reusable or depends on your implicit knowledge.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[Exercises/Workflow Automation/Wa Advanced 01|WA-Advanced-01]] — where you'll design and document a complete AI-automated workflow for a business process.

Back to [[Workflow Automation|Workflow Automation]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Workflow Automation Exercises]]

The Multi-Source Brief

2026-03-16T16:23:12Z

Admin: Fix 1 internal link(s)

''Synthesize outputs from three separate AI perspectives into a single coherent analysis. 25 minutes.''

<blockquote>
'''One-liner:''' Synthesize outputs from three separate AI queries into a single coherent analysis — building the skill of triangulating AI perspectives.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a question or topic you need to actually understand — a market trend, a technology choice, a strategic decision, a complex issue in your field.

'''Run three separate queries''' in three different AI sessions (or clear context between each). Each query approaches the same topic from a different angle:

'''Query 1 — The Optimist:'''
<blockquote>
Analyze '''[your topic]''' from the most optimistic perspective. What's the strongest case that this will succeed/matter/grow? Cite specific evidence, trends, and examples. Be persuasive, not balanced.
</blockquote>

'''Query 2 — The Skeptic:'''
<blockquote>
Analyze '''[your topic]''' from a skeptical perspective. What's the strongest case that this is overhyped, risky, or likely to fail? Cite specific evidence, counterexamples, and historical parallels where similar things didn't pan out. Be rigorous, not cynical.
</blockquote>

'''Query 3 — The Analyst:'''
<blockquote>
Analyze '''[your topic]''' by identifying the 3-5 key variables that will determine the outcome. Don't argue for or against — map the decision space. For each variable, describe what would need to be true for a positive outcome vs. a negative one. Include what we don't yet know.
</blockquote>

'''Now synthesize.''' Open a fresh document (not an AI chat). Write a 250-word brief that answers:
# '''What do all three perspectives agree on?''' (This is likely true.)
# '''Where do the Optimist and Skeptic directly contradict each other?''' (This is where the real uncertainty lives.)
# '''Which of the Analyst's key variables would resolve the contradiction?''' (This is what you need to investigate.)
# '''Your take''' — Given all three inputs, what's your position and what would change your mind?

The brief should be something you'd share with a colleague or decision-maker. No AI jargon, no meta-commentary about the process.

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a topic''' — Pick something with genuine uncertainty. If the answer is obvious, the exercise won't stretch you. Good candidates: emerging trends, strategic choices, technology bets, or contested ideas in your field.
# '''Run three separate AI sessions''' — Optimist, Skeptic, and Analyst. Use fresh contexts (new chats or cleared conversations) so each query isn't influenced by the others.
# '''Read all three outputs''' — Don't start synthesizing until you've read all three. Notice your own bias — which perspective did you instinctively agree with?
# '''Write the synthesis yourself''' — This is the critical step. Don't ask AI to synthesize for you. The skill you're building is ''your'' ability to integrate contradictory information.
# '''Distill to 250 words''' — Force compression. A good brief is one where every sentence earns its place.

'''"Done" looks like:''' A 250-word brief you'd be comfortable sharing with a colleague, built from three distinct AI perspectives, with a clear statement of what you believe and what would change your mind.

----

== 🧭 Why this matters (Strategists start here) ==

In [[Exercises/Insight Synthesis/Is Basic 01|IS-Basic-01]], you extracted insights from a single AI output. Here, you're building a fundamentally harder skill: '''triangulating across multiple AI perspectives to form your own judgment'''. This is exactly what senior decision-makers do with human advisors — they don't take any single perspective at face value. The discipline of writing the synthesis yourself (rather than asking AI to do it) ensures you're developing the judgment, not outsourcing it. This skill directly applies to research, due diligence, competitive analysis, and any situation where multiple data sources tell different stories.

----

== Reflection ==
* Which perspective (Optimist, Skeptic, Analyst) was most useful? Which felt like filler?
* Did writing the synthesis yourself change your view compared to where you started? At what point in the writing did it shift?
* Would you share this brief with a decision-maker? If not, what's missing?
* 💬 ''Share your 250-word brief with someone who knows the topic. Ask them what they'd challenge — their pushback will tell you where your synthesis was weakest.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[Exercises/Insight Synthesis/Is Advanced 01|IS-Advanced-01]] — where you'll build a full research synthesis pipeline with structured evidence evaluation.

Back to [[Insight Synthesis|Insight Synthesis]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Insight Synthesis Exercises]]

The Handoff Protocol

2026-03-16T16:23:11Z

Admin: Fix 1 internal link(s)

''Split a problem across two separate AI sessions with different roles, then synthesize their outputs. 25 minutes.''

<blockquote>
'''One-liner:''' Split a problem across two separate AI sessions with different roles and contexts, then synthesize their outputs yourself — like managing a real team.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

You'll need two AI chat windows open at the same time (two browser tabs, or two different AI tools — either works).

'''Pick a project or decision''' that has at least two distinct dimensions. For example: "Create a content strategy for launching our new product."

'''Chat A — The Strategist.''' Open your first chat and send:

<blockquote>
You are a '''brand strategist''' with 15 years of experience. Your focus is positioning, audience targeting, and messaging clarity. You do NOT think about implementation details — that's someone else's job.

I'm working on: '''[your project]'''

Give me your strategic recommendations. Focus on: who the audience is, what the core message should be, and how to position this differently from competitors. Be specific and opinionated.
</blockquote>

'''Chat B — The Executor.''' Open your second chat and send:

<blockquote>
You are an '''operations-focused content producer'''. Your focus is practical execution: channels, formats, timelines, and resource requirements. You do NOT set strategy — you receive it and figure out how to make it real.

I'm working on: '''[your project]'''

Give me an execution plan. Focus on: which channels to prioritize, what content formats work best, a realistic timeline, and what resources I'll need. Be specific and practical.
</blockquote>

'''Now you're the manager.''' Read both outputs. Notice what Chat A assumed that Chat B would question, and vice versa. Then write your own synthesis:
* Where do these perspectives align?
* Where do they conflict?
* What did each one miss that the other caught?
* What's your actual plan, informed by both?

'''Optional bonus round:''' Take your synthesis and paste it back into one of the chats:

<blockquote>
Here's the combined strategy and execution plan I've built from two different advisors. Poke holes in it. What's still weak?
</blockquote>

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Choose a project''' — Something real with both a strategic and practical dimension. Content launches, product decisions, event planning, and hiring processes all work well.
# '''Set up Chat A (Strategist)''' — Give it a clear strategic role with explicit boundaries. Tell it ''not'' to worry about implementation.
# '''Set up Chat B (Executor)''' — Give it a clear operational role with explicit boundaries. Tell it ''not'' to set strategy.
# '''Run both chats''' — Send the same project description to each, but with their respective role prompts.
# '''Synthesize manually''' — You are the integration point. Compare outputs, find gaps, resolve conflicts, and produce a combined plan.

'''"Done" looks like:''' You have a plan that neither AI session could have produced alone, and you can articulate what each perspective contributed.

----

== 🧭 Why this matters (Strategists start here) ==

In [[Exercises/Agent Collaboration/Ac Basic 01|AC-Basic-01]], you simulated multiple perspectives in a single chat. Here, you're practicing a fundamentally different skill: '''managing separate agents with isolated contexts'''. This mirrors how real multi-agent systems work — each agent has a specific role, limited scope, and doesn't see the other's work. The human (you) acts as the orchestrator. This is the skill that scales: from two chats to entire AI-assisted workflows with specialized roles, handoff points, and quality gates.

----

== Reflection ==
* How did the outputs differ when each AI had a constrained role vs. a single AI doing both? Was the split worth the extra effort?
* What context got lost in the handoff between sessions? How would you design a better transfer summary?
* Did the synthesis step feel harder or easier than you expected? What made it difficult?
* 💬 ''Run this exercise with two colleagues, each managing one AI session. Compare the experience of synthesizing someone else's AI output vs. your own — it highlights how much implicit context lives in your head.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[Exercises/Agent Collaboration/Ac Advanced 01|AC-Advanced-01]] — where you'll design a complete multi-agent workflow with defined roles, handoffs, and feedback loops.

Back to [[Agent Collaboration|Agent Collaboration]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Agent Collaboration Exercises]]

The Framework Transplant

2026-03-16T16:23:10Z

Admin: Fix 1 internal link(s)

''Systematically transplant a problem-solving framework from another domain to solve your challenge. 25 minutes.''

<blockquote>
'''One-liner:''' Take a complete problem-solving framework from another domain and systematically adapt it to solve a challenge in your own work.
</blockquote>

----

== 🔧 Jump in (Tinkerers start here) ==

Pick a challenge you're currently facing in your work — something you've been approaching the same way without breakthrough results.

'''Step 1 — Find a foreign framework.''' Send this prompt:

<blockquote>
I'm struggling with '''[your challenge]''' in my field of '''[your field]'''. I want a completely fresh approach. Give me 3 well-known problem-solving frameworks from '''different''' fields (engineering, medicine, military strategy, game design, ecology — anything outside my domain). For each framework:
1. Name and origin field
2. How it works (3-4 step process)
3. Why it might apply to my problem

Choose frameworks that are genuinely different from each other, not variations on the same idea.
</blockquote>

'''Step 2 — Deep-dive one framework.''' Pick the most promising or most surprising framework. Send:

<blockquote>
Let's go deeper on '''[chosen framework]'''. Walk me through how a professional in '''[its origin field]''' would apply this framework to a real problem in their domain. Be specific — give me a concrete example with actual steps, not abstractions.
</blockquote>

'''Step 3 — Systematic transplant.''' Now adapt it:

<blockquote>
Now help me transplant this framework to my challenge: '''[restate your challenge]'''.

Map each step of the framework to my context:
- '''Step 1 of framework''' → What does this look like in my situation?
- '''Step 2 of framework''' → What's the equivalent action?
- (continue for all steps)

For each mapping:
- What translates directly?
- What needs to be modified and how?
- What doesn't transfer at all, and what should replace it?

End with a concrete action plan I can execute this week.
</blockquote>

'''Step 4 — Stress test.''' Send:

<blockquote>
Play devil's advocate. Where does this transplanted framework break down when applied to my field? What assumptions from the original domain don't hold in mine? How should I adjust?
</blockquote>

----

== 📋 Plan first (Planners start here) ==

Here's what you're about to do:
# '''Identify your challenge''' — Pick something real where your current approaches have stalled. The exercise only works if you're genuinely stuck.
# '''Discover foreign frameworks''' — Use AI to surface structured problem-solving approaches from unfamiliar fields. Look for frameworks with clear steps, not just theories.
# '''Study the framework in its native context''' — Understand how it actually works in practice before trying to adapt it. This prevents shallow borrowing.
# '''Map step-by-step to your context''' — Systematically translate each step, noting where the mapping is direct, where it needs modification, and where it fails entirely.
# '''Stress test the adaptation''' — Identify where the transplant breaks down and adjust before committing to action.

'''"Done" looks like:''' A concrete action plan for your challenge, based on a framework from another field, with clear documentation of what translated, what was modified, and what was replaced.

----

== 🧭 Why this matters (Strategists start here) ==

In [[Exercises/Cross Domain Reframing/Cdr Basic 01|CDR-Basic-01]], you borrowed a single technique from another field. Here, you're transplanting an '''entire framework''' — a much harder and more valuable skill. This is how breakthrough innovations happen: the structure of a solution transfers across domains even when the details don't. Toyota's production system was adapted from supermarket inventory management. Agile software development borrowed from lean manufacturing. The ability to systematically adapt frameworks across domains is what separates insight from coincidence. At the advanced level, you'll build an entire cross-domain prompt library; this exercise builds the adaptation methodology.

----

== Reflection ==
* Which parts of the framework transferred most easily? What does that tell you about the underlying structure of your problem?
* Where did the transplant break down? Was the breakdown due to domain differences, or did it reveal an assumption you hadn't questioned?
* Did the stress test change your action plan significantly, or just refine the edges?
* 💬 ''Explain the transplanted framework to someone in the original field. Their reaction ("that's not how we use it" or "interesting adaptation") tells you whether you captured the core principle or just the surface.'' (Social Learners)

== ⬆️ Level up ==

Ready for more? Try [[Exercises/Cross Domain Reframing/Cdr Advanced 01|CDR-Advanced-01]] — where you'll build a cross-domain prompt library with documented transfer patterns.

Back to [[Cross-Domain Reframing|Cross-Domain Reframing]]

[[Category:AI Fluency Playbook]]
[[Category:Exercises]]
[[Category:Cross-Domain Reframing Exercises]]