Thinking in Roles

Module 3, Multi-Agent & Delegation | Essay 2 of 7

Before you can delegate to multiple agents, you need to be able to decompose a task into distinct roles. This is the foundational skill of multi-agent work, and it's more intuitive than it sounds.

A role is a focused responsibility with clear boundaries. The researcher finds and synthesizes information. The writer turns that information into a draft. The editor improves the draft without rewriting it from scratch. The reviewer checks the whole thing against the original goal. Each role has a different focus, different standards, and a different definition of "done."

The way to identify roles in a task is to ask: what are the distinct phases of this work, and what mindset does each one require?

Research requires breadth and skepticism, gather as much as possible, evaluate sources, surface what matters. Writing requires narrative judgment, what's the story, what's the structure, what's the right level of detail. Editing requires precision, what's redundant, what's unclear, what breaks the flow. Review requires distance, does this actually accomplish what it was supposed to?

These mindsets conflict. The best researchers are often expansive; the best editors are ruthless. The best writers make confident choices; the best reviewers question everything. Putting all of those in one agent at once produces tension that dilutes each phase.

Role boundaries matter too. An agent with a clearly defined role does better work because it's not second-guessing itself. A writer who knows someone else is doing the research doesn't pad the draft with extra context. An editor who knows the goals upfront doesn't try to reimagine the structure. Clear roles let each agent do its job cleanly.

You don't need elaborate role designs for everyday tasks. Two roles, produce and review, covers a lot of ground. Research plus write is another simple split that improves most information-based work. Start simple, add roles where you see quality gaps.