What is a one-person AI agency? A plain definition

A dog groomer in your town wants a new website. Ten years ago she might have gone to an agency with a designer, a developer, someone to write the words and someone else to answer her emails. Today she might hire one person who does all of that with an AI agent beside them.

She may never know the difference, and she shouldn't have to. What she'll notice is whether the site arrives when you said it would, whether the booking link works, and whether her emails get a reply.

That one person is running what people have started calling a one-person AI agency. This article explains what the phrase means, what the agent does, and what still needs you.

What is a one-person AI agency?

A one-person AI agency is a business run by one owner that sells services to other businesses, usually starting with websites, where an AI agent does much of the making. The agent writes code, drafts copy, researches and checks. The owner picks the clients, agrees the price, decides what's good enough and answers for the result.

Each word in the name does a job:

• Agency, because the client pays for an outcome, such as a working website and someone who looks after it, and one person is responsible for it.
• One-person, because there is no team behind the owner. No account manager, no junior designer, no office.
• AI, because much of the work a team used to split between them is now shared between the owner and a program.

It helps to say what it isn't, too. It isn't a company that sells an AI product. It isn't a way to have software run a business while you do something else. And it doesn't remove the parts of client work that need a person: agreeing what's included, taking responsibility, and being reachable when something breaks.

What does the AI agent actually do?

An AI agent is a program that takes a task, works through the steps, and uses tools along the way, such as reading files, running commands or searching. Anthropic describes its agent, Claude Code, as an AI coding assistant that helps you build features, fix bugs and automate development tasks, and that can work across many files and tools in a project.

For a small web business, that usually means the agent:

• builds pages from a brief you wrote;
• drafts first versions of page copy for you to edit;
• gathers what's public about a business before you pitch it: its site, its listing, how customers find it;
• checks the work, such as broken links, forms that don't send, and how pages look on a phone;
• turns notes from a call into a scope or a list of tasks.

Agents can also read from the programs you already use. In November 2024 Anthropic released MCP, short for Model Context Protocol, a shared way to connect AI programs to your files and other tools. With a connection like that, your agent can work from your client notes instead of whatever you last pasted into a chat.

Here's how the work tends to split on a typical website job:

The job	What the agent does	What you do
Research before a pitch	Gathers public facts and says where it found each one	Checks the facts and decides whether to pitch
The quote	Drafts it from your notes	Sets the price and sends it
The build	Writes the pages and fixes what you point out	Reviews, asks for changes, picks the version that goes live
Client emails	Drafts a reply	Reads it, edits it, sends it
Monthly care	Runs the checks and lists what it found	Decides what to fix and what to tell the client
Money	Drafts the invoice	Checks it and confirms the payment arrived

What do you still do yourself?

More than the excitement around AI suggests. The agent can make things. It can't be accountable for them, and clients hire a person for exactly that.

These stay with you:

1. Choosing who to work with. Which businesses to approach, and which jobs to turn down.
2. Agreeing price and scope. What's included, what's extra, and what happens when the client asks for more.
3. Taste. Deciding what's good, what fits this business and what its customers need.
4. Approving what goes out. Anything a client or the public will see should wait for you to read it. Make approving what your agent sends a habit from the first job, before there's a reason to regret skipping it.
5. Keeping promises. Remembering that you said Friday, that the café wanted its winter menu up, that the dentist's form goes to the wrong inbox. Keeping track of clients and promises is most of the job once you have more than two clients.
6. Money. Quotes, deposits, invoices and checking that a payment actually arrived.
7. Being reachable. Answering when a client's site is down, even when the agent isn't running.

If that list looks like the business side of an agency, that's because it is. The agent takes on much of the making. The running stays human.

Who does a one-person AI agency suit?

It suits people who can judge client work and would rather not do every part of it by hand:

• designers who know what a good site looks like and want help with the build;
• developers who'd rather spend their time on the hard parts than on the fifth contact page this month;
• marketers who already sell services to local businesses and want to add websites;
• people starting a one-person AI agency, perhaps next to a day job, who need to look organised from the first client.

It suits people less well if they want to avoid clients altogether. The agent doesn't change the fact that someone has to talk to the bakery owner about why the site needs her opening hours.

What does it take to start one?

Less than a traditional agency, and more than signing up for an AI subscription. A sensible first month looks like this:

1. Pick one service. A website with a monthly care plan is the usual start. Add other services later, one at a time.
2. Pick a small patch. One suburb or one kind of business, so your research and examples build on each other.
3. Set up your own accounts in your own name. Email, payments, your AI subscription. Client accounts, such as their domain, belong in the client's name.
4. Write down your process. How you research, what you check before a client sees anything, what's in a care plan. Your agent works better from written rules, and so do you.
5. Decide your approval rule. Nothing reaches a client until you've read it. Write it down so you keep to it on busy days.
6. Keep one record per client. Who they are, what you agreed, what you promised, what they've paid.

None of this promises clients or income. It gives you a business that holds together when the first client says yes.

What are the common mistakes?

A few come up again and again when one person takes on agency work with an agent:

• Letting the agent send things unread. A reply with the wrong price or the wrong client's name costs more trust than the time you saved.
• Quoting on the agent's speed. A page that took the agent minutes still needs your review, the client's feedback and a round of changes. Price your time and the value to the client, not the build alone.
• Keeping client details in chats. A chat window is a poor filing cabinet. When the chat ends, the context goes with it.
• Holding the client's accounts yourself. If the domain is in your name and you get sick, the client is stuck. Set accounts up in their name from the start.
• Taking every kind of job. One person can't be a web shop, an SEO team and a social media department in the same month. Say no to work that doesn't fit your service.

A one-person AI agency works when the agent does the making and you keep hold of the decisions, the promises and the money. Start with one service, one patch of town and one written rule: nothing goes out until you've read it.