An AI agent for project management, when you are the whole team

An AI agent for project management sounds like a robot project manager. That is not the useful version. When you run your own projects, you are the project manager. The agent is the one collaborator you have: it takes a defined piece of work, does it, and reports back in a place you can both see.

The agent is not your project manager

Most AI tools for project management are built for teams. They summarise meetings, draft status reports, and suggest who should take which task. That is useful when there are twenty people and a steering group. When there is one of you, there is nobody to report to and nobody to assign work to. Except the agent.

So turn the question around. Not "can an AI run my projects?" but "which parts of my projects can I hand to an agent, and where does it tell me how it went?" You keep the decisions: what a project is for, what matters this week, and what counts as finished. The agent gets work.

What it looks like in a shared project

Picture one of your projects. A website, a client job, a house you are slowly fixing up. It has a list of tickets, and each ticket is one problem or one piece of work with a short title. This is what an AI agent for project management looks like in practice:

1. You write a ticket, or ask the agent to file one when it finds something.
2. The agent picks up the ticket you point it at and marks it in progress.
3. It does the work and writes in the ticket's thread what it did and what it found.
4. If it needs a decision, it marks the ticket blocked with one concrete question, and waits.
5. When the work is finished, it marks the ticket done, with a note.

You read the thread when it suits you. On your phone, on the train, next week. You reply, close the ticket, or send it back. The project remembers everything, so neither of you has to.

What it takes off your plate

• The first draft. Research, a comparison, an outline, a list of options with the trade-offs spelled out.
• The tedious pass. Checking every page of a site for broken links. Going through a folder of quotes and pulling out the prices.
• Filing what it notices. One ticket per problem, instead of one long message you will never finish reading.
• Following up. Picking a ticket up again in a new session and reading its own notes first.

None of this is magic. It is ordinary work, handed over with enough context to be done, and handed back with enough detail to be checked.

What stays with you

An agent can be fast, confident and wrong, all at once. So some things should never leave your hands:

• Choosing. Which projects exist, and which one matters now.
• Defining done. What a finished ticket looks like, before the work starts.
• Anything with consequences. Spending money, contacting people in your name, or anything that cannot be undone.
• Saying it is live. Done and deployed are different things. The agent can finish the work. You, or a check you trust, confirm it is out.

For the longer version of where to draw these lines, read human in the loop, when you are the only human.

Why a team tool fits badly

Team tools are good at what they are for: many people, roles, permissions, workload charts and reports that travel upward. Bring one person and one agent into that and you get empty fields, seats you pay for, and settings for meetings that will never happen. The agent also tends to get access to far more than the one project it needs.

Why a chat window fits badly

A chat is great for a quick answer and a poor place to keep a project. The conversation grows, scrolls away, or gets cleared. Nothing has a status. "What did we decide about the header?" turns into an archaeology project. And when you open a new chat, the agent starts from nothing.

What to look for in a place you share with an agent

• Scoped access. The agent gets one project, not your whole life, and you can cut it off at any time.
• A trail. Every change and every status change has a name on it, so you can tell your notes from the agent's.
• Statuses that mean something. At least new, in progress, blocked and done. Ideally a separate one for confirmed live.
• Small tickets. A title and a few lines. If a ticket needs a page of explanation, it is probably a project.
• Readable from anywhere. You check progress where you are, not only at the desk where the agent runs.
• Cheap and quiet. It should not need maintaining.

This is still personal project management. The only difference is that the list is shared with one helper who can actually do some of the work.

How to start

Pick one project. Write three small, clear tickets. Connect the agent to that project and nothing else. Tell it three rules: one ticket per problem, blocked with a question when it needs you, done with a note when it is finished. Read what comes back, and fix the tickets before you blame the agent. If your agent is Claude, connecting it to one project takes a minute.

Very Simple Projects is a simple personal project management app built around exactly this: projects, numbered tickets, a thread on every ticket, and six statuses you can add to. You connect an agent to one project with its own token, which you can revoke at any time. It files tickets, reports in the thread, and marks a ticket deployed only when the change is confirmed live. Every status change lands in the thread with its name, and it cannot delete comments or see projects you did not give it. There are no team seats. It costs $0.99 a month, and the first month is free.