Linear published aggregated product data from its paid workspaces showing that AI-created issues have nearly reached parity with human-created ones: 2.435 million per week against 2.481 million, measured in early August 2026. Two years ago, fewer than one issue in a thousand was created by AI.
The other headline number
Pull requests per workspace are up 111% since June 2024. Teams connected to coding agents were opening roughly 65 PRs per team per week as of June 2026, against about 10 for teams without them.
What the common telling gets wrong
The 65-versus-10 comparison will be quoted as a productivity multiple. Linear says directly that it is not one: those teams "were already higher-output before coding agents existed, so the levels aren't directly comparable." It is a selection effect. What the data can support is each cohort's trajectory against its own baseline — and on that measure the agent-connected cohort did accelerate.
The second caution is Linear's own, and it is the most useful sentence in the report: "an opened PR says nothing about the value of the change." Issues created and pull requests opened are measures of motion. Neither counts whether anything merged, worked, or should have been built. A metric that rises when software writes more tickets for itself is not self-evidently a measure of output.
What the parity does and does not show
Issue creation is a low-cost act, and it is the act agents are best positioned to perform in volume — filing follow-ups, splitting work, logging findings. Parity in issues filed is not parity in work done, and reading it that way would be the error the publisher of the data explicitly warns against. The scope is also bounded: paid Linear workspaces and the users in them, which skews toward software teams that adopt tooling early.
Why it is still worth knowing
Going from under 0.1% to roughly half of everything created in a tracker in two years is a real change in who or what populates the queue engineers work from — regardless of whether each item is worth doing.