AI for Project Managers: Twelve Jobs It Can Take Off Your Plate

AI is most useful to a project manager as a first-draft machine for the paperwork around the work: status updates, meeting actions, risk lists, backlog cleanup, dependency questions, stakeholder notes and release notes. In each case the assistant reads material you give it and writes a draft; you check the facts, the tone and anything that commits the team. What you keep entirely are commitments to clients, decisions about people, and money. The twelve jobs below each say what the AI drafts and what you check. The wider picture of how assistants fit into a team is in what is AI project management?, and the prompts themselves are in prompts for project management with AI.

The rule behind all twelve

An assistant writes confidently whether or not it knows. NIST’s Generative AI Profile↗ (opens in a new tab) (NIST AI 600-1) calls this confabulation: false content stated as if it were true. So every job below follows the same pattern. Give the assistant the source material, tell it to mark anything it cannot find as unknown, and read the result before anyone else does. The time you save is in the writing, not in the checking.

Twelve jobs, and what you check on each

1. Weekly status drafts. The AI drafts: done this week, next week and needs a decision, from the board and your notes. You check: every item is real, nothing red is buried, and the audience is right.
2. Meeting notes into tasks. The AI drafts: a list of actions, each with a verb, an owner and what done looks like. Microsoft documents that Copilot in Teams meetings↗ (opens in a new tab) suggests action items, and Google Meet’s notes do the same. You check: the owner agreed, and nothing said in passing became a commitment.
3. Risk lists. The AI drafts: risks read out of the plan, with a likelihood, an impact and a first response for each. You check: the ones only you know about, such as a key person leaving or a client under pressure.
4. Backlog grooming. The AI drafts: likely duplicates, stale tasks nobody has touched in months, and tasks too big to finish in one go. You check: nothing a customer is waiting on gets closed as stale.
5. Dependency questions. The AI drafts: which tasks wait on others, and the questions to ask each owner. You check: the answers, by asking the people.
6. Stakeholder updates. The AI drafts: one source rewritten for three audiences, such as the sponsor, the client and the team. You check: nothing the client should not see, and no promise you have not made.
7. Release notes. The AI drafts: user-facing notes from the tasks marked done, grouped as new, improved and fixed. You check: each item actually shipped, and the wording makes sense to a customer.
8. Kickoff and charter drafts. The AI drafts: goals, scope, out of scope and open questions, from a brief. You check: scope matches what was sold.
9. Agendas. The AI drafts: an agenda built from open questions and blocked tasks. You check: the order, and who needs to be in the room.
10. Estimates and task breakdowns. The AI drafts: a first cut of a feature split into smaller tasks. You check: the sizes with the people doing the work, not with the model.
11. Retrospective summaries. The AI drafts: themes grouped from the team’s notes. You check: that quiet voices and awkward points survived the summary.
12. Decision logs. The AI drafts: a record of what was decided, why and what was turned down, from a meeting or a thread. You check: a person made the decision, and the record names them.

Most of these start from the same short prompt: the source, the audience, the length, and an instruction not to fill gaps. Sixteen of them, ready to paste, are in prompts for project management with AI. A way of deciding which of your own jobs to hand over, and which to keep, is in which tasks can AI agents automate.

What never to hand over

• Commitments. Dates, scope and anything a client will hold you to. An assistant can draft the email; a person decides what it promises and sends it.
• People decisions. Hiring, performance, who is on which project, who gets blamed for a slip. These need judgment, context and accountability that a model does not have.
• Budgets and money. Approving spend, changing a budget line, accepting a quote. An assistant can reconcile invoices against milestones and list the differences; a person signs.
• Anything irreversible. Deleting a project, closing a client’s tasks, sending to a mailing list. If it cannot be undone, a person presses the button.

Writing these down helps more than remembering them. Put them where every assistant reads them before it starts, not in each person’s chat history.

Why the work belongs on a board, not in a chat

Every job above is better when the assistant reads the same current list of work that the team does. A status draft built from last week’s export is last week’s status. Meeting actions left in a chat are lost to everyone who was not in it. And if one person’s assistant files a task and another person’s assistant closes it, you want to know which did what.

That is what a shared board gives you: one list of work that people and assistants both read and write. Assistants reach it over the Model Context Protocol↗ (opens in a new tab), the open standard Claude, ChatGPT, Copilot, Gemini CLI and others use to connect to tools. The assistant reads the board, drafts what you asked for, files what came out of the meeting, and the changes are there for everyone the next morning. If you are new to the idea, what is MCP explains it in a page.

How fenbs fits

fenbs is a task board where people and AI assistants are members with roles. Tasks move through four lanes, To Do, Next Up, In Progress and Completed, and each is a feature, an enhancement or a bug with a priority from 1 to 10, where 1 is the most urgent. A task holds a note for the problem, a plan for how it will be done, and a test status with notes. An assistant connects to https://fenbs.ai/api/mcp, signs in through the browser, and reads the team’s rules from the Decisions and rules page before anything else, then the AI context notes. That is where the list above belongs: “a person approves anything that commits the client” as a rule, recorded by the person who decided it.

For the jobs above, Add many turns a pasted list of meeting actions into tasks, and Copy as Markdown copies the board for an assistant that is not connected. History records who changed what, with an assistant’s changes shown under its own name. fenbs has no due dates, sprints or assignee field, so an owner or a date goes in the first line of the task note, and there are no Gantt charts or budgets: those stay in the tools you use for them.