Founder Playbook

Agent Personas

Use named agent roles to make autonomous systems legible, trainable, and easier to manage.

Agent personas turn a pile of prompts into an operating team. Instead of one vague assistant doing everything badly, founders define clear roles, behaviors, and quality bars for each agent.

Why this matters

  • They make delegation concrete. A founder can say, "Researcher, gather evidence" and "Editor, tighten the narrative."
  • They improve reliability because each agent has a narrower job, tighter instructions, and clearer acceptance criteria.
  • They make collaboration easier for humans. People remember characters and roles faster than prompt files.

How founders can use it

  • Start with 3 roles max: researcher, operator, editor. Do not create a cast of 12 on day one.
  • Give each agent one mission, one style, one output format, and one escalation rule.
  • Measure each persona on one hard metric, such as response quality, cycle time, or error rate.
  • Document handoffs between personas so work moves cleanly instead of looping.

Failure modes to watch

  • Cute names do not fix bad system design. If the tools, context, or workflow are weak, the persona will still fail.
  • Avoid overlapping roles. Two agents with the same job create confusion, not leverage.
  • Do not let persona theater replace accountability. A named agent still needs tests and checklists.

Operator questions

  • Which 3 recurring jobs consume founder attention every week?
  • Where does a specialist agent beat a generalist one?
  • What exact output should each persona own?

Referenced in

Founder takeaway

Do not treat this concept as trivia. Use it to sharpen a decision, redesign a workflow, or find a better wedge into the market.