What is AI project management?
It is not a chatbot bolted onto a task list. It is a board where an AI assistant is a named member with a role, and everything it changes is signed.
7 min read
AI project management is project management in which AI assistants do part of the work on the plan itself: they read the tasks, pick one up, change it, and write down what they did. In practice it only works when the assistant is treated like any other member of the board. It has an identity, it holds a role that limits what it can do, and every change it makes has its name against it. Anything less is a chat window sitting next to a task list.
That distinction is the whole subject, so the rest of this piece explains it: what changes when an assistant stops advising on your work and starts working on it, and what a board needs in order to make that safe.
A chat assistant and a board member are different things
Most people first meet AI in project work as a chat assistant. You paste in a list of tasks, ask for a summary or a plan, and copy the answer back into whatever tool you use. The assistant never touches the real list. It cannot, because it has no way in, and so the question of what it is allowed to do never comes up.
The moment you connect an assistant to the real board, that question is the only one that matters. It can now move a task to Completed, rewrite what a task says, or add ten new ones. If the board cannot tell you who did that, or stop it from happening, you have a problem that no amount of clever summarising makes up for.
- A chat assistant works on a copy. A board member works on the one real list that everybody else is looking at.
- A chat assistant has no identity. A board member has a name that appears in the history.
- A chat assistant has no limits because it has no access. A board member has a role, and the role is the limit.
- A chat assistant forgets. A board keeps the task, the notes and the record of what was done, whichever assistant you use next.
What it looks like day to day
On fenbs the loop is short. Somebody writes a task in one line, for example “Search ignores the category filter”. It gets a ref such as BU-016, and it sits in one of four lanes: To Do, Next Up, In Progress or Completed. Later, at a desk, you tell the AI assistant you already use to take the next thing. It reads the task, works out the detail, does the job, moves the card, and records in plain language what it actually did. You read that before you have opened the code.
None of that needs a special AI feature inside the board. The assistant is Claude, ChatGPT, Cursor or whatever you use, connected over a standard protocol called MCP. The board’s job is smaller and more important: to be the single place the work lives, and to be strict about who may change it. The how it works page walks through the five steps.
fenbs_whoami → Editor on fenbs.ai fenbs_list_items → 4 tasks
The first call matters. fenbs_whoami tells the assistant who it is acting as and what it may do on the board. An assistant that cannot find out what it is allowed to do will guess, and guessing is how it loses your trust in the first five minutes.
The three things that make an assistant accountable
1. An identity you can name and take away
On fenbs you connect an assistant by issuing it a token under Settings, in the section called Connect an AI assistant. You give it a name, such as “Claude” or “the nightly triage script”. The token is shown once and cannot be read back. Revoking it takes one click, stops it at once, and does not sign you out. That is the difference between an assistant with its own key and an assistant borrowing your password.
2. A role that limits it
A token acts as the person who issued it and inherits their role on the board. On top of that it carries scopes: read, write and comment. Both are checked, and the narrower one wins, so an assistant can never do something its owner could not. If your own role is narrowed, every token you issued narrows with it, immediately. There is no separate bot-permissions screen to keep in step with the real one.
3. A name against every change
Every change on the board is recorded with who made it, and each task shows its own trail. An assistant’s changes are recorded by name, in the form “Claude via” the person it acts for, so its work can always be told apart from a person’s. Revoke the token and that record stays. Taking away access should not erase what happened.
Being told no is part of the design
When an assistant tries something outside its role, the refusal is a full sentence naming the permission that was missing and the role it holds: “You are a Viewer on Clerkenwell Showroom, which cannot move tasks between lanes.” The assistant can repeat that to you instead of retrying blindly or quietly doing something nearby.
What AI project management is not
It helps to be plain about the limits. Connecting an assistant to your board does not plan your product for you, and it does not remove the need for someone to decide what matters. The assistant works through the list; a person still owns the list.
It is also not a reason to add machinery. fenbs has no epics, sprints, story points, custom fields or automations, and that is deliberate. An assistant does not need a complicated board in order to be useful. It needs a clear one: a task with a title and a note, a lane that says where the work is, and a kind that says whether it is a Feature, an Enhancement or a Bug. The simpler the model, the less there is for a person or an assistant to get wrong.
Questions to ask of any tool
If you are comparing tools for a team that uses AI assistants, these are fair questions to put to every one of them, including ours.
- Can I tell, for any change, whether a person or an assistant made it, and which one?
- Can I give an assistant less access than I have myself, for example read only?
- Can I take an assistant’s access away without changing my own password or signing myself out?
- Does the record of what it did survive after I take its access away?
- When the assistant is refused, is it told why in words it can pass on to me?
- If I change to a different assistant next month, does the work stay where it is?
On fenbs the answer to each is yes, and the connection guide shows the setup. Other tools answer these questions in their own ways, and some teams will need things fenbs leaves out on purpose. The questions are the useful part.
Where to start
Start with one board and one assistant. Write a handful of real tasks, connect the assistant with read access only, and ask it what is on the board. When you are comfortable, issue a token that can also write and comment, and ask it to take the next task. Read what it recorded. If the trail makes sense to you without opening the code, the setup is working. AI assistants cost nothing extra on any plan; the details are on the pricing page.