About KirinBoard

Why we built it, and how it works.

AI coding agents changed how fast software gets written. A change that took a week can take an afternoon, and a team can run many agents at once. The rest of the process did not speed up with it. Every change still has to be reviewed, tested, merged and released, and those steps still move at human speed.

So the hard part moved. It is no longer writing the code. It is getting many streams of work to come together quickly without breaking each other:

• Two agents pick up the same task from different prompts and fix it two different ways.
• Two branches change the same files, and nobody finds out until the merge.
• Pull requests arrive faster than anyone can review them, then queue for test servers.
• Changes are merged and counted as done, but never reach your users.
• Each agent works in its own terminal, so nobody sees the whole picture.

The question every team ends up asking is: How do I keep a firm grip on my project when many agents are working on it at once?

We wrote KirinBoard to coordinate our own agents, and started sharing it with friends and companies who were experiencing similar issues. As KirinBoard became central to their agent coordination, productization of KirinBoard was the next logical step.

Agents keep the record as they work. Our agent pack is a short instruction file your agents read, plus a small script. It asks every agent to do four things:

• File the work under an initiative. Every task is a GitHub issue, attached as a sub-issue of a larger initiative (what Agile calls an epic), so you can always answer "what is this for?"
• Say what stage it is in. One stage label per task (scoping, planning, building, end-to-end testing, review, deploying, monitoring), moved on as the work moves. A "blocked" flag sits beside it when the agent is stuck, so you see where the work stopped.
• Say what it holds. Before it starts, the agent posts a claim on the issue: this session holds this task and expects to change these files. It keeps the claim fresh while it works.
• Connect the pull request to its task, so the change, its tests and the task show up together.

KirinBoard reads that record back to you. The Delivery board groups your work by initiative: what is in each stage, what is blocked or has gone quiet, which initiatives are almost done, and what has shipped. Because GitHub timestamps every label change, it can replay progress over any period you choose, not just today. The Agents board shows who holds what, and warns when two agents or two branches are about to change the same files. The CI board shows your tests and test servers. KirinBoard also scores how much of your work sits under an initiative and carries a stage, so you can see where the habit is slipping, and truth checks flag where the labels and what really happened disagree. It can only look, never change anything.

This is Agile-lite. We kept the parts of Agile that make work visible: a backlog, epics (initiatives), board columns (stages), and a strict definition of done, where merged is not done until the change is live and behaving. We left out the parts that exist to coordinate people's calendars. Agents work in minutes and hours, not two-week sprints, so there are no story points and no stand-ups. We measure flow instead: how fast work moves, and where it waits. Nothing is enforced unless you choose to enforce it.

One thing we decided not to do is build another tracking tool. GitHub issues, labels and sub-issues stay the record, so if you stop using KirinBoard, your work still makes sense.

If you are looking for a tool to help coordinate and orchestrate your agents, then go ahead: sign up and connect your first project!

KirinBoard is a product of Kirin Solutions LLC.