AI NewsProductivityAnnouncement
Linear's own engineers merged 53 percent of their pull requests with no second human review in October
Linear published its own code-review figures for 2026. PRs merged with no second human reviewer went from 1 percent in January to 53 percent in October.

Why it mattersA team that measured code review by review count now has a point of comparison. Linear's own engineers moved past the second human on half of merges.
A code review number from inside Linear now reads as a warning or a promise. The reader decides which. In October 2026, 53 percent of the merges on Linear's own repository went through with no second human reviewer. In January 2026, it was 1 percent.
Linear published those figures on 16 November 2026 in the second edition of its data series. The authors are Tim Qi, Lukas Eipert, Maciek Pękala and Dima Zaytsev. The data is Linear's own code history, counted monthly across the ten months of 2026.
The headline numbers
Pull requests merged per month went from 2,111 in January to 5,341 in October. That is a 153 percent increase. Over the same period the number of humans writing those pull requests grew by 22 percent, from 58 to 71. The post reports 72 merged PRs per human author in October, against 36 in January.
The share merged with no second human reviewer tells a sharper story. One percent in January. Fourteen percent in July. Fifty-three percent in October. The authors note that the median PR size also grew. It went from 47 to 108 lines of code.
What rose at the same time
The post adds a second number that is harder to celebrate. Pull requests closed without being merged went from 251 in October 2025 to 1,182 in October 2026. That is roughly five times more. The authors tie the rise to agent-written code. The team opens the pull request, looks at it, and closes it.
Review requests per author fell too. June: 85. October: 54. Engineers were reviewing less. The authors read the fall as a shift in where attention goes: away from review count, toward merging decisions.
What the data does not say
The post is one team's own numbers, from a company that makes project-management software. The authors do not claim the ratio applies to other teams. They do not report the share of PRs authored by an agent as a separate figure. The reader cannot see the split in those 5,341 merges. How many came from a human, and how many came from an agent in a loop.
The method is defensible. Every merge on the team's own repository is counted. Tenure, role and PR size are reported. The gap is that Linear sells a tool many of the readers use at work. A team measuring the same ratio in a different industry may see something else.
What to count next
A team that tracks code review by requests per author now has a second column worth tracking. The share of merges with no second human review. The share rose past half at Linear over a year. A lead whose team sees the same climb has a simple choice. Trust the first reviewer and the tests, or slow the merge rate back down. The 1,182 closed-not-merged PRs say the slower path is not free either.
Source
Code review at Linear, edition 02 by Tim Qi, Lukas Eipert, Maciek Pękala and Dima Zaytsev.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.


