The agent roles a freelancer actually runs

A one-person firm does not need a hundred agents; it needs five well-defined ones, each owning a single job, because a role with one clear remit is the only kind an agent reliably gets right.

Before you wire anything, you decide the roster. Hiring is the same discipline whether the staff are people or agents: one job per box, no overlap, and a clear answer to "who owns this when it breaks". The roster below is the concrete org chart a freelancer actually runs, drawn from the work that fills most service weeks.

The five core roles

A useful fleet starts narrow, with the five roles that recur across nearly every freelance practice.

1. The researcher. Gathers the raw facts a piece of work needs: market context, competitor moves, source material, the brief behind the brief. Its output is a tidy, sourced brief, never a finished deliverable. It feeds the drafter.
2. The notetaker. Captures every call and meeting so nothing lives only in your memory. Its output is a transcript plus a short action list, which becomes input for both the drafter and the scheduler.
3. The drafter. Turns research and notes into a first version of the actual deliverable: the proposal, the client update, the article, the email. It is the workhorse, and it never sends; it hands to the reviewer.
4. The scheduler. Owns your calendar and the calendar of the work: it books calls, blocks focus time, and sequences tasks so the fleet's output lands when it should.
5. The reviewer (QA). Reads the drafter's output against a fixed standard and decides whether it is good enough to reach a human. This is the role that protects your reputation, and it is the one most freelancers skip.

Each role is one job. The researcher does not draft; the drafter does not send; the reviewer does not write. That separation is what makes the chain debuggable later, because when an output is wrong you can name the box that owns the failure.

Why one job per agent, and not one super-agent

It is tempting to build a single agent that "handles a client end to end", and it is the wrong move.

A super-agent has no seam you can inspect. When its output is wrong you cannot tell whether the research was thin, the draft was sloppy, or nobody checked it, because all three happened in one opaque pass. Splitting the work into roles gives you a seam at every hand-off, so you can test each stage in isolation and improve the weak one without touching the rest. Specialised agents also hold tighter instructions: a drafter prompt that only ever drafts stays sharper than one prompt asked to research, draft, schedule, and review at once.

The cost is a few more moving parts, and it is worth paying. You are trading one fragile genius for five reliable specialists, which is exactly the trade a growing firm makes when it stops relying on one heroic generalist.

Two roles you can staff today with proven tools

Two of the five roles map onto mature, off-the-shelf agents you can put to work this week, before you build anything custom.

The notetaker role is solved by Fireflies: it joins your calls, transcribes them, and produces a structured summary with action items, which is precisely the notetaker's output contract. You do not build a transcription agent from scratch; you adopt one and treat its output as the feed into your drafter and scheduler.

The scheduler role maps onto Motion, which takes your tasks and meetings and auto-arranges them into a defended daily plan. Rather than hand-rolling calendar logic, you let the scheduler agent own the calendar and feed it the action items your notetaker produces. Adopting proven agents for solved roles is not a shortcut; it is the right build order, because it lets you spend your custom-build effort on the researcher, drafter, and reviewer that carry your specific craft.

A worked example: the roster that unblocked a four-client cap

Take a freelance content marketer who kept dropping balls between calls and deliverables.

• Challenge. Notes lived in three notebooks, research happened in browser tabs that vanished, and drafts were written from memory the night before they were due. Quality depended entirely on how tired she was.
• Approach. She named the five roles explicitly and staffed the two easy ones first: Fireflies as the notetaker on every call, Motion as the scheduler turning each call's action items into booked focus blocks. With notes and scheduling no longer in her head, she had a clean feed for the drafter she built next.
• Result. Action items captured per call went from "whatever I remembered" to every item, and missed internal deadlines dropped from about three a week to zero within a fortnight. The roster gave each task an owner, so nothing fell through the gaps between boxes.

Pitfalls when defining the roster

• Overlapping remits. If two agents could both "write the update", neither owns it and both drift. Give every job exactly one owner.
• Skipping the reviewer. A roster without a QA box ships the drafter's first attempt straight to clients. The reviewer is not optional; it is the role chapter five is built around.
• Building before adopting. Hand-rolling a notetaker or scheduler when Fireflies and Motion already do the job wastes the effort you should spend on your craft roles.

INTERVIEW EWOUD: Which agent role in your own fleet earns its keep most, and which one did you underestimate until a job slipped through because no agent owned it?

With the five boxes named and two of them already staffed, the next question is plumbing: how each agent reaches your apps and data, and what makes it fire. The next chapter covers wiring the fleet.