The One-Person Company Playbook: Roles, Agents, Review Loops

A one-person company with AI agents works when you treat the agents as a junior team: each one owns a clearly defined job, drafts the work, and hands it to you for approval before anything reaches a customer. You stay the owner of decisions, money and relationships. The agents take the repetitive production work off your plate.

That is the whole playbook in two sentences. The rest of this guide is how to set it up: which roles to hand off first, how to write an agent's job so it does the job well, and the daily review loop that keeps quality up.

What a one-person company actually is

A one-person company is a business with one owner and no employees. That's not a fringe category. The U.S. Census Bureau counted 29.8 million nonemployer businesses in 2022, businesses with no paid staff. Most of them are consultants, freelancers, shops and service providers, not startups chasing a billion-dollar valuation.

What changed in the last two years is the ceiling. A solo owner used to hit a wall where growth meant hiring: someone to write, someone to post, someone to answer inquiries. Agents push that wall back. They don't remove it. You can now get a lot of production work drafted for you, but someone still has to decide what is good enough to ship. In a one-person company, that someone is you.

The well-known version of the idea is Sam Altman's 2023 remark that tech CEOs had a betting pool on when the first one-person billion-dollar company would appear. That's an interesting headline, but it's the wrong benchmark for most readers. The useful question is narrower: which hours of your week can an agent draft, and which still need you?

The four roles in a one-person company

Every small business, solo or not, runs four kinds of work. Sorting your tasks into these buckets tells you where agents help and where they don't.

The pattern: agents fit the producer, responder and researcher roles. The owner role stays human. The more a task touches money, commitments or someone's feelings, the more it needs you.

Which roles to hand off first

Don't automate everything at once. Hand off one role, run it for two weeks, then add the next. Here's a sensible order for most service and content businesses.

1. The researcher

Research is the safest first handoff because a bad output costs you nothing but a few minutes. A research agent can read competitor pages, summarize news in your niche, or build a one-page brief before a sales call. You read it and decide what matters. A research or chief-of-staff style agent is a low-risk way to learn how agents think and where they go wrong.

2. The content producer

Content is the job most solo founders fall behind on, and the one agents help most with. A writer agent that knows your business, your customers and your voice can draft a weekly article or a batch of posts. You edit and approve. See how a content writer agent is typically set up.

3. The responder

Review replies and first responses to inquiries are next. They're repetitive and they're time-sensitive, and a good draft is most of the work. They're also customer-facing, so the review step matters most here. A review response agent or a lead response agent should always draft, never send.

What to keep

Keep these with yourself, even if a tool offers to do them: setting prices and discounts, refunds and credits, contracts and commitments, hiring and firing, anything legal or medical, and the first conversation with a big client. These aren't hard tasks for an agent to attempt. They're hard to undo when the agent gets them wrong.

There's good evidence for drawing the line here. When Anthropic and Andon Labs let a Claude agent run a small office shop in Project Vend, the agent was talked into discounts, sold items below cost, and lost money. In the second phase, procedures and checklists made it much more reliable, but staff still found ways to get free items. Money decisions need a human in the loop.

How to write an agent's job so it works

Most disappointing agents have a vague job. "Help with marketing" produces generic output. A good agent brief reads like a job description for a new hire. Use this template:

Name: Maya, content writer
Job: Draft one blog post per week for [business], aimed at [customer type]
        who are trying to [problem].
Inputs: Our website, the services list, the voice guide, last month's posts.
Output: A 1,200-1,800 word draft with a title, 3 subheads, and one
        practical checklist. Saved for my review.
Rules:
  - Never invent statistics, prices, customers or quotes.
  - Use US English, short paragraphs, second person.
  - Don't mention competitors by name.
  - If you're unsure of a fact, leave a [CHECK] note instead of guessing.
Done means: I can publish it after under 15 minutes of edits.

Three parts do most of the work:

1. A narrow job. One agent, one recurring output. If you want a writer and a social manager, make two agents.
2. Shared knowledge. Every agent should work from the same facts about your business: your site, your services, your prices, your voice. Contradictions between agents usually come from each one having different context.
3. A "done" definition. Tell the agent what good looks like in a way you can check. "Publishable after 15 minutes of edits" is a target you can measure.

The review loop: the part that makes it work

The review loop is the difference between a one-person company and a one-person mess. The rule is simple: agents draft, you decide (here's a deeper guide to setting up an AI agent approval workflow). Nothing goes to a customer, a website or a social account without passing through you.

The daily loop

1. Agents run on a schedule. Research every weekday morning, content drafts weekly, review replies as reviews come in.
2. Drafts land in one place. An inbox, a folder, a queue. One place, so you're not hunting across tools.
3. You review in a fixed block. Same time each day, 20 to 45 minutes. Approve, edit or decline each item.
4. You turn edits into rules. If you fix the same thing twice, write it down as a rule the agent keeps. "Never promise same-day service." "Sign off as the team, not a name."
5. You check the log weekly. What ran, what you approved, what you declined and why.

Step 4 is where the time savings come from. In week one you'll edit a lot. By week six the drafts should need far less, because the agent has your rules.

A review checklist you can copy

Before approving any agent draft, check:

• Every number, price and date is correct and sourced.
• No promise you can't keep (timelines, guarantees, discounts).
• It sounds like you, not like a press release.
• No names, customers or quotes that weren't in the source material.
• It answers the actual question or does the actual job.
• You'd be comfortable if a customer screenshot it.

A sample week for a solo consultant

Here's how the roles and the loop fit together for a hypothetical one-person marketing consultancy. These are illustrative times, not measured results; your numbers will differ.

The founder still works. The work just moves from writing first drafts to client work, sales and judgment calls.

Common ways one-person AI setups fail

• Too many agents too soon. Ten agents you don't review is worse than two you do. Add one at a time.
• No shared facts. Each agent pulls from different notes and they contradict each other. Put the facts in one place.
• Auto-publishing. The fastest way to an embarrassing post or an angry customer. Keep the approval step even when drafts get good.
• Treating review as optional. Benchmarks still show agents failing most multi-step office work on their own; in Carnegie Mellon's TheAgentCompany benchmark, the best model tested fully completed about 30% of tasks. Review isn't overhead. It's the job.
• Measuring hours, not outcomes. Track what shipped (articles published, inquiries answered within the hour), not how busy the agents look.

For a deeper look at where agent-run setups break, read can a business be run by AI agents. If you're weighing agents against hiring help, AI agent vs virtual assistant compares cost and quality side by side.

Tools for running the playbook

You can build this loop with general tools: a chat assistant for drafts, a shared doc for your facts, a calendar reminder for the review block. That works, but you become the glue, copying context into every conversation.

A dedicated agent platform keeps the pieces together. Full Circle is built around this exact loop. You hire agents from templates (content writer, social media manager, lead intake, research briefer, review responder) or describe your own. Every agent shares a company brain built from your website and the documents you upload, plus a voice guide built from your writing samples. Agents run on schedules, can brief each other, and send every finished piece to an approval inbox where you approve, edit or decline. Nothing sends or publishes on its own. Agents also keep the rules you give them in chat. Plans start at $39 a month for one agent; see pricing or browse agent use cases.

Whatever you use, the playbook stays the same: narrow jobs, shared facts, a daily review block, and edits turned into rules. If you're thinking of selling agent work to others instead of running your own company with it, see how to start an AI agent business.

The short version

• Sort your work into owner, producer, responder and researcher.
• Hand off research first, then content, then responses. Keep money and commitments.
• Write each agent's job like a job description, with rules and a "done" definition.
• Review every draft in a fixed daily block, and turn repeated edits into rules.
• Add one agent at a time, and judge them by what ships.