AI Now Writes Half of All Linear Issues

Here is a statistic that would have sounded absurd two years ago. On the project-tracking platform Linear, artificial intelligence now writes just under half of all the issues teams create, up from fewer than one in a thousand in 2024. Coding agents have tripled the number of pull requests some teams open each week. And yet, by Linear's own reading of the data, those teams are shipping a little slower, not faster.

That last part is the interesting bit, and it is easy to miss under the headline number.

More output, not more done

It is tempting to read "half of all issues are written by AI" as proof that the machines have taken over the backlog. The reality is more mundane and more useful. An issue is a request for work, not the work itself. Agents are very good at generating them: spotting a flaky test, filing a bug, drafting a plan, opening a pull request. Turning that flood of activity into shipped, reviewed, trustworthy software is still the hard part, and that is where the extra time is going.

This lines up with something we looked at recently: the gap between how productive AI coding tools feel and how much they actually move the needle. More pull requests is not the same as more finished features. Every agent-written issue still needs a human to triage it, and every agent-written pull request still needs someone to review it. Review does not get faster just because the code arrived quicker.

The tracker becomes the control room

Linear's response has been to lean in rather than pull back. The company has rebuilt issue tracking as the place where you assign, review and approve work done by agents, with cloud coding sessions that run on Claude Code and Codex, and author labels that flag which text a machine wrote. There are even "Loops," recurring agent workflows for chores like weekly triage.

It is a sensible bet. If agents are going to generate half the tickets, the tracker has to become the place where humans stay in charge of the queue. Whether that finally tips the math from "more activity" to "more shipped" is the question the next year will answer. For now, Linear's numbers are a clean illustration of a lesson the industry keeps relearning: agents are good at starting work, and finishing is still hard.