Project managers

How project managers use AI agents

Project management is the work of turning ambiguity into a plan and then keeping everyone honest about it. A kickoff lands as a pile of half-formed notes that has to become scope, milestones, and a date. Every week needs a status update that tells the truth without setting off a panic. Risks need naming before they bite, dependencies need mapping before they tangle, and the meeting that just ended needs to become action items with owners before everyone forgets what they agreed to. None of it is the project. All of it is what keeps the project from quietly going sideways.

An AI agent is good at this exact layer because so much of it is structured text and arithmetic built on things you already track. It turns messy kickoff notes into a plan with milestones and surfaced assumptions. It pulls action items and decisions out of pasted notes. It does honest timeline and resourcing math instead of optimistic guessing, so a deadline is something you can defend. And it drafts the status update and the slip-the-date message in a calm, direct voice that flags bad news early instead of burying it.

Wire it to Slack and you can fire a blocker update from your phone between meetings. Save your status format as a Skill and every week's update comes out in the same clear shape. Put your project facts and your working assumptions in Memory once, and the prompts below cover the plan, the comms, and the math. You run the project. The agent does the writing-up and the arithmetic.

Open the Agent10 min read

Capabilities this leans on

Calculator File upload Memory Skills Scheduled tasks Connections

Set up Memory once

Do this first. Every plan, update, and risk list below comes back in your voice and against your real working assumptions.

Remember these facts about my work: I'm a project manager at Northstar, a B2B software company, running cross-functional product and marketing projects. My teams are engineering, design, marketing, and a contracted QA vendor. We work in two-week sprints, track in Jira and Linear, and report status to a weekly leadership sync on Mondays. My writing voice is clear, calm, and direct: I flag risks early and never bury bad news. Standard working assumptions are 6 productive hours per person per day, Fridays are light, and I always separate facts from asks in an update.

1.Turn a messy kickoff into a plan

Hand it the raw notes; get scope, milestones, and the questions you still need answered.

Here are my raw kickoff notes for the Q3 billing-revamp project, pasting below. Turn them into a structured project plan: the goal in one sentence, scope and explicit out-of-scope, the major workstreams, a milestone list with rough sequencing, and the open questions I still need answered before we commit a date. [paste notes]

From that plan, propose 5 milestones with target weeks assuming a kickoff this Monday and a hard deadline of September 30, and tell me which milestone is the riskiest to hit.

List the assumptions you made so I can correct any that are wrong.

What you get: A structured plan with milestones and surfaced assumptions, built from notes that were a mess an hour ago.

2.Pull action items out of meeting notes

Paste or upload the notes and get owners, due dates, and decisions, cleanly separated.

Here are the notes from today's billing-revamp sync (attached). Pull out every action item as owner, task, and due date. Flag any task with no clear owner, and list the decisions we made separately from the actions.

Now write a 5-line recap I can post in the project channel, leading with the decisions, then the actions, then anything still open.

What you get: A clean action list with owners and a channel-ready recap, so nothing agreed in the room gets lost.

3.Draft the status update you keep rewriting

Give it the status and a few bullets; get a structured update and a Skill to repeat it.

Write this week's status update for the billing-revamp project for the Monday leadership sync. Status is yellow: design is on track, engineering is one sprint behind on the migration, QA starts next week. Structure it as overall status, what shipped, what's next, risks, and asks. Clear and direct, no burying the slip.

Save this as a Skill called 'Status update' so I can generate one any week by giving you the project, the RAG status, and a few bullets.

What you get: A truthful, well-structured update in two minutes, and a saved Skill that keeps every week's update in the same shape.

4.Build the risk and dependency list

Name the risks and map the critical path before either one surprises you.

For the billing-revamp project, draft a risk register: at least 8 risks across technical, resourcing, scope, and external-vendor categories. For each give likelihood, impact, an early warning sign, and a mitigation. Be specific to a payments migration, not generic.

Now map the dependencies: which workstreams block which, where the critical path runs, and the single dependency that would hurt most if it slips.

What you get: A specific risk register and a dependency map that show you where the project is actually fragile.

5.Write the stakeholder comms, including the bad news

Deliver a slip honestly without spin and without spooking the team.

Engineering hit a blocker and the billing-revamp launch is going to slip two weeks. Write the message to leadership that delivers the slip honestly: what happened, the new date, the impact, and what we're doing about it. Calm, accountable, no spin. Under 150 words.

Now a shorter, warmer version for the wider project channel that keeps the team steady and clear on what changes for them.

What you get: Two versions of a hard message that own the slip and keep trust intact, drafted before you talk yourself out of sending it.

6.Do the timeline and resourcing math

Get a defensible date from real numbers instead of optimistic guessing.

The billing-revamp build is estimated at 240 engineering hours. I have 2 engineers at 6 productive hours a day, 5 days a week, but one is out for 4 days this sprint. How many calendar days to finish, and what's the earliest realistic completion date from a Monday start? Show the math.

If leadership wants it done a week sooner, what are my options: add a third engineer, cut scope, or extend hours? Quantify what each one would take and flag the risk in each.

What you get: A completion date you can defend in the room and a quantified read on every option to pull it in.

7.Run the Monday digest from Slack

Scheduled tasks and a connected channel keep status moving without you at a desk.

Every Monday at 7am, draft me a status digest across my active projects: for each, the RAG status, what moved last week, what's due this week, and any new risk. Ask me for the inputs you don't have.

(From Slack) Quick one: turn these three bullets into a clean blocker update for the channel, lead with the ask. [paste bullets]

What you get: A status digest that drafts itself before the week starts, and clean updates you can fire from Slack between meetings.

Run your first prompt

Open the Agent, paste any prompt above, and change the details to fit your business.