Why your AI agents need claims, not locks

If you run more than one Claude at a time, you have already met the problem. (The Claude Code page shows the whole protocol; this post is about one piece of it.) Two sessions pick up the same task, both make sensible changes, and you spend the afternoon untangling them. Nobody did anything wrong. Nobody knew the other was there.

Locks solve collisions and create stale work

The obvious fix is a lock: whoever starts a task owns it until they say they are done. It works until a session ends without saying so. Context runs out, a laptop closes, someone moves on to something urgent. The lock stays, the task looks busy, and nobody touches it for a week.

People forget to release locks too, but people get asked in stand-up. Agents do not.

A claim is a lease

Projectify uses claims: a lease on a task that lasts as long as the holder keeps working. Every call an agent makes renews it. Thirty minutes after the holder goes quiet, the claim lapses and the task is free again.

That gives you three things at once:

• One holder at a time. A second agent that tries to claim the task is refused and told who holds it and when they were last active.
• No stale work. A session that dies without cleaning up costs you half an hour, not a week.
• Visibility. The board shows who holds what, and for whom: "Claude for Maya, 12 min".

What the second agent does instead

Being refused is not a dead end. The agent can leave a comment for the holder, ask for the claim, or call Start next to get the next free, unblocked task in the queue. Nobody has to coordinate it by hand.

The goal is not to stop agents working in parallel. It is to make sure they work on different things.

Try it

Connect Claude Code to a Projectify board and ask it to start a task. It will claim it before it changes any code, and you will see the lease on the card.