Editing Agent Collaboration (section)

== What this looks like at each level ==

=== Basic — Role-based prompting ===

You're learning to give AI a defined role and perspective, rather than asking it as a generic assistant. The core skill: framing AI as a specific expert and comparing perspectives to make better decisions.

'''What it feels like:''' You run your first "AI team meeting" — asking AI to respond as two different experts on the same problem. You notice that the dual-role output is more nuanced than a single generic answer. You start using role-based prompts by default.

=== Intermediate — Managing separate agents ===

You've moved from simulating multiple roles in one chat to actually running separate AI sessions with isolated contexts. Each session has a specific role, limited scope, and doesn't see the other's work. You act as the orchestrator — the person who manages the handoff and synthesizes the results.

'''What it feels like:''' You run two AI chats in parallel (strategist and executor), manually pass information between them, and produce a plan that neither could have generated alone. You notice where context gets lost in handoffs and design better transfer summaries.

=== Advanced — Agent architecture ===

You're designing complete multi-agent workflows with defined roles, inputs, outputs, handoff triggers, and feedback loops. You think about agent architecture before writing prompts — deciding how to decompose work, what each agent needs to know, and where human checkpoints belong.

'''What it feels like:''' You design a 4-agent pipeline (researcher, drafter, critic, editor), run it end-to-end on a real project, and produce output that's measurably better than a single "do everything" prompt. You can explain ''why'' you split the work the way you did.