What is an AI project manager?
An AI project manager is software that does the upkeep a project manager would do for a software team. It reads where the work happens (meetings, Slack, GitHub, Linear) and keeps the official record matching what the team actually decided. It drafts the issue, the status change, or the follow-up, and a person approves it.
The difference from other AI tools is that it doesn’t wait to be asked. A planning call ends and the draft issues are already waiting. A ticket goes quiet for three weeks and it says so. It watches for the gap between what people agreed and what the tracker says, then proposes the fix.
It helps to be clear about what it isn’t:
- Not the AI built into your tracker. That AI is strong inside one tool. An AI project manager works across them, because most drift happens between tools: the decision lands in Slack, and the ticket still says the old thing.
- Not a general chat assistant. A chat assistant answers from whatever you paste in, and nothing gets written down as decided. An AI project manager reads the work itself and keeps one record.
- Not a meeting summarizer. A summary tells you what was said. An AI project manager turns what was said into owned, tracked work, and follows up when it stalls.
- Not a coding agent. It doesn’t write code. It keeps the plan and the tickets honest, so the people and coding agents who do write code start from what was actually decided.
Thor is an AI project manager for software teams. The engine underneath, one current record of what the team decided, is described on the context layer page.
Why are software teams adopting AI project managers now?
Because writing code got faster and coordinating the work didn’t. Coding agents open pull requests faster than anyone can keep the board current, so the plan in the tracker falls behind the plan in people’s heads.
Three things make that gap expensive:
- Many teams have no dedicated project manager. The upkeep lands on an engineering manager, a tech lead, or the one product manager covering three teams. It gets done on Thursday afternoon, from memory, by re-reading Slack.
- Decisions are scattered. A planning call changes the scope. The change is agreed in a Slack thread. The ticket, the doc, and the roadmap each still say something different.
- Stale tickets now mislead machines too. A coding agent briefed from an out-of-date ticket builds the wrong thing with total confidence. We wrote about the real cost of stale context and why agents confidently do the wrong thing.
An AI project manager takes the upkeep off people’s desks, so the person who used to do it by hand spends the time deciding instead.
What does an AI project manager do across a sprint?
It drafts the coordination work at each step of the sprint: issues from planning calls, ticket cleanup in the backlog, standups and recaps during the week, and roadmap checks. People keep the decisions. Here’s how that splits for each job, and what Thor does today.
AI sprint planning
The job: turn a planning conversation into a set of owned, tracked issues, and keep them current when the plan moves.
What AI can do: read the planning call transcript, draft the issues with owners and a link back to where each was agreed, and draft a move to next sprint when a Slack thread changes the plan.
What stays human: sprint scope, capacity, and priority. Deciding what fits is a judgment call about people and trade-offs.
What Thor does today: after a Google Meet or Zoom call, Thor reads the transcript and drafts the GitHub or Linear issues. You approve, Thor files them. When a later Slack decision changes the plan, Thor drafts the update to the issue, such as moving it to the next sprint. Thor doesn’t set sprint scope or capacity. More on this in what happens after a meeting ends and turning a rough idea into a task breakdown.
AI backlog grooming
The job: keep the backlog honest. Close what already shipped, merge duplicates, and catch tickets that stopped matching reality.
What AI can do: spot tickets that shipped, tickets that repeat each other, and tickets that went stale, then draft the cleanup.
What stays human: whether a ticket still matters. “Stale” and “parked on purpose” look the same from the outside.
What Thor does today: Thor finds the tickets that already shipped, repeat each other, or went stale, and drafts the cleanup. You approve, Thor closes them. It also flags tickets that have gone quiet compared with what’s normal for that kind of work, rather than applying one fixed day count. See finding the ticket nobody’s touched in three weeks and which one is true, the ticket or the Slack thread.
Standups and status updates
The job: let everyone know what moved, what’s blocked, and what’s next, without a meeting or a hand-written report.
What AI can do: write each person’s standup from their actual activity, and write the weekly recap from what closed, what’s blocked, and what has no owner.
What stays human: the context behind a blocker, and what to do about it.
What Thor does today: Thor sends a daily standup summary in Slack or Microsoft Teams, a manager summary in Slack, and a Friday wrap-up and Monday kickoff each week. You can also just ask Thor “what’s my standup?” These go out without approval because they report, they don’t change your tools. Our post on the weekly status report nobody reads covers why this beats the hand-written version.
Roadmap checks
The job: make sure the work in flight still serves the roadmap.
What AI can do: compare open issues against the roadmap and flag the ones that don’t fit.
What stays human: what stays on the roadmap and what gets cut.
What Thor does today: Thor checks open issues against the roadmap and flags what doesn’t fit. You decide what stays.
Estimation, risk, and blockers
The job: know how big the work is and whether it will land.
What Thor does today: Thor doesn’t estimate effort or story points, and it doesn’t give a numeric risk score or a delivery forecast. It flags the signals that come before a missed date: stalled threads, aging commitments, tickets that drifted from what was said, and issues that are blocked or have no owner. A person reads those and makes the call.
Manual vs. AI-assisted: who does what?
AI handles the reading and the drafting. A person still approves changes and makes the calls about scope, priority, and people. This table shows the split for common project management jobs.
| Job | Done by hand | With an AI project manager | Still needs a person |
|---|---|---|---|
| Planning call to issues | Someone types up issues from notes after the call | Issues drafted from the transcript, with owners and a source link | Approving the issues, and deciding scope |
| Plan changes in Slack | Someone remembers to update the ticket, or doesn’t | The ticket update is drafted from the thread | Approving the update |
| Backlog cleanup | A quarterly grooming session | Shipped, duplicate, and stale tickets found and cleanup drafted | Whether a ticket still matters |
| Stalled work | Spotted when someone asks “what happened to X?” | Flagged when it goes quiet compared with normal | Deciding what to do about it |
| Standups | A daily meeting or a typed update | Written from each person’s activity | Adding context behind blockers |
| Weekly status | A hand-written report on Friday | A recap of what closed, what’s blocked, and what has no owner | Reading it and acting on it |
| Roadmap fit | Checked at planning time, if at all | Open issues checked against the roadmap | What stays and what’s cut |
| Estimation | Planning poker or a lead’s guess | Not covered by Thor | All of it |
How does an AI project manager compare with other approaches?
There are five common ways a software team handles project management upkeep, and each stops somewhere different. The fair question is which job you need done. Thor makes this guide, so read the Thor row with that in mind.
| Approach | What it’s good at | Where it stops |
|---|---|---|
| Manual (a person plus a tracker) | Full judgment and context | Doesn’t scale, and the upkeep is the first thing dropped in a busy week |
| AI built into one tracker | Summaries and drafting inside that tool | Doesn’t see Slack, meetings, or the other tools where decisions happen |
| General AI chat assistant | Answering a question from what you paste in | Waits to be asked, keeps no shared record, and two people can get two answers |
| Standalone AI work-management app | Planning and scheduling inside its own app | Usually means moving your work into a new tool your team has to keep current |
| AI project manager across your tools (Thor) | Keeping GitHub, Linear, Slack, and meetings telling the same story, with a person approving changes | Doesn’t write code, estimate effort, or decide priorities. Works with GitHub and Linear issues today, and reads Notion and Google Docs without editing them |
Where Thor fits
Thor comes to the work instead of asking your team to come to it. Slack or Microsoft Teams stays your chat, GitHub or Linear stays your tracker, Notion or Google Docs stays your docs. Thor reads meeting transcripts from Google Meet and Zoom, and it never joins the call as a bot.
When something should change, Thor drafts the change, shows where it came from, and waits for a person to approve it before anything in your tools changes. We explain why in propose, then approve.
Coding agents such as Cursor and Claude Code can read the same record through Thor MCP, so they start from what the team decided rather than from a stale ticket. If you’re weighing Thor against tools that call themselves AI teammates, our comparison with AI-teammate tools goes through each one.
Who uses an AI project manager?
Whoever absorbed the coordination work without choosing to. On a software team that’s usually an engineering manager, a product manager, or, in a smaller company, the CTO.
Engineering managers and tech leads
You’re the person who ends up routing information: chasing status, turning calls into tickets, noticing what slipped. Thor drafts that upkeep and flags stalled work, so your time goes to the team instead of the board. See Thor for engineering teams.
Product managers
You’re often covering more teams than one person can track. Thor turns planning calls and Slack decisions into issues, keeps tickets current when the plan changes, and checks open work against the roadmap. See Thor for product teams.
CTOs and engineering leaders
You want to know where things actually stand without asking five people. Thor’s weekly recaps and stalled-work flags give you that from the work itself, not from a status meeting. See Thor for engineering leaders.
How do you roll out an AI project manager?
Start small, keep a person approving every change, and measure against a baseline you took before you started. Here’s a six-step rollout that works for a team of ten or more.
- Pick one team and one tracker. Start with a single team on GitHub or Linear, not the whole company.
- Connect chat and the tracker first. Chat plus a tracker is enough to be useful after your next standup. Add meetings and docs once the team trusts the drafts.
- Write down your baseline. Before you turn anything on, note how long weekly status takes, how many tickets have gone quiet, and how long a planning call takes to become tickets. Without a before, you can’t tell whether it worked.
- Keep approval on. Every change to a ticket waits for a person. Read the drafts for the first two weeks and correct what’s wrong. That’s also how the team learns to trust it.
- Turn on standups and weekly recaps. These report rather than change your tools, so they’re a low-risk way to see what Thor sees.
- Measure after a month, then widen. Compare against your baseline. If the upkeep went down and the tracker matches reality more often, add the next team.
Setup takes a few minutes and doesn’t need a workspace admin for most connections. You can start from Get Started.
FAQ
What does an AI project manager do?
It does the upkeep a project manager would do for a software team. It turns meetings and Slack decisions into issues, keeps tickets and boards current, flags stalled work, writes standups and weekly recaps, and checks open issues against the roadmap. It drafts changes, and a person approves them.
Can AI replace a project manager?
No. An AI project manager takes over the upkeep: the typing, the chasing, and the checking. Scope, priorities, trade-offs, and the conversations with people stay with a person. Teams without a dedicated project manager get the upkeep done without hiring for it, and teams with one get that person’s time back for the judgment calls.
Will an AI project manager change my tickets without asking?
Not with Thor. Thor drafts the change, shows where it came from, and waits for a person to approve it before anything in GitHub or Linear changes. Standups, weekly recaps, and nudges on stalled threads go out on their own, because they report on the work rather than change it.
Do I have to leave my current tracker?
No. Thor works inside the tools you already use. Slack or Microsoft Teams stays your chat, GitHub or Linear stays your tracker, and Notion or Google Docs stays your docs. Thor reads meeting transcripts from Google Meet and Zoom.
How is an AI project manager different from ChatGPT or the AI in my tracker?
A general assistant answers from what you paste into it and waits to be asked. The AI in a tracker is strong inside that one tool. An AI project manager reads the work across your tools, keeps one shared record of what was decided, and comes to you when a commitment is aging or a ticket no longer matches what the team agreed.
Is it safe to give an AI project manager access to our tools?
Thor is SOC 2 Type II compliant and has completed a CASA assessment. It never trains models on customer data, conversations, or meeting transcripts. For GitHub, Linear, Google, and Notion, Thor uses each person’s own authorization, so it can’t see what that person can’t see. It can create, update, and comment on issues, but it can’t push code.
Does Thor estimate story points or forecast delivery dates?
No. Thor doesn’t estimate effort or give a risk score or delivery forecast. It flags the signals that come before a missed date: stalled threads, aging commitments, drifting tickets, and blocked or unowned issues.
How much does an AI project manager cost?
Thor’s Free plan is $0 with 25 actions a month and no credit card. Team is $199 a month with 300 actions, and $0.45 for each action after that. Enterprise is custom. Every plan includes unlimited people, meetings, and connected tools. An action is a finished thing Thor produces, like a ticket created or a question answered. Standups and nudges on stalled work don’t count.