Project Management for AI Coding: Why Your Board Goes Stale

On Monday you set up a board: twelve cards, a tidy plan for two weeks. Your AI finishes the first card in twenty minutes. Then, because it was already in there, it reworks the login flow, renames two database tables and fixes a bug you had not written down yet.

By Thursday the board says four things are in progress. Two are done. One was abandoned on Tuesday. The fourth stopped making sense after the rename. Nobody lied to the board. It just stopped being told anything.

Why the board falls behind

Most project management tools assume the person doing the work is the person updating the tracker, at human speed. A developer finishes a ticket, drags it to done, writes a line about what changed. A minute of admin against a day of work.

With an AI writing the code, that ratio flips. The work takes minutes. The update still takes a minute, and now it is your minute, spent describing changes you watched scroll past. So you save it for later, and by Friday the board describes a project that no longer exists.

Underneath that sits a second gap. The board lives in a tab; your AI lives in the chat. Unless something carries the plan across, every session starts from whatever you remember to paste in.

What people try, and where it breaks

Discipline. Update the board after every session. It holds for a few days and gives way on the busiest ones, the days the record matters most.

Connect the tracker to the AI. Linear, for one, offers a hosted connector that lets an AI client find, create and update its issues. That closes the visibility gap, and it is a real improvement. But an assistant that can update the board still waits to be asked, and nothing in a normal session asks unless you do. Which is the discipline fix again.

A plan file in the repo. A task list in a markdown file, or a small command-line tracker the AI is told to check at the start of each session. This is closest to the right shape, because the AI reads it where it works. It breaks more quietly. Files describe. They don't decide. Nothing checks the finished work against them, and they never reorder the backlog when reality changes.

The plan has to sit in the path of the work

Project management for AI coding works when keeping the plan current is a side effect of building, rather than a chore you do afterwards. The plan is the input to every session, and the record is its output.

Before a build, the AI reads the task: the goal, what is out of scope, how you will tell it is finished. After the build, it writes back what actually happened, including changes that were never on the card. You check the result against the task and decide whether it stands. The next plan starts from those write-ups, so Monday's rename is already in Tuesday's plan.

A board that only you update is only ever as current as your last free hour.

The other half is letting the plan change. Your idea will grow while you build it, and that is the work going well. A tracker that calls every change of direction slippage teaches you to stop recording changes. Write the new decision down with its reason and mark the old one replaced.

One way to run it

This is the job we built PAPI for. PAPI is a project manager that works inside your AI chat.

Work runs in cycles, usually 3 to 8 tasks. Each task gets a build handoff with the goal, what is out of scope, how to check it is done, the risks and the likely files, written so an AI with no history on the project can run it. Every build files a report: actual effort against the estimate, surprises, issues found on the way. You accept, request changes or reject, and the next plan reads those reports first. Decisions keep their reasoning, and a changed mind supersedes the old one rather than overwriting it.

PAPI isn't an app you open. It works where you work. Your assistant writes and reads the board as part of the build, so nobody reconstructs the week on Friday. We build PAPI with PAPI, and have for several hundred cycles.

When to stay where you are

If your team already runs on Linear or Jira and the board is current, you do not need to move. Different category; keep Linear if your team lives there. Our Linear comparison sets out where each one fits, including using both.

If you are still exploring and throwing away most of what you build each week, a plan is overhead. Wait until the project holds something you would be upset to lose.

And PAPI does not write code. Your assistant still builds, and the call on whether the work is good enough stays with you.

Try it for one week

Whatever tool you use, run one week with two rules: no task starts without a written check for done, and no task closes without a note on what actually changed. If the board is still true on Friday, you have the habit that matters. The longer version of that loop is in Plan, Build, Review. Walk one cycle on a demo project, no signup needed.