An agent that opens a pull request with no human owner produces three pull requests where one survives. That is the number I would put in front of anyone building a software factory.
LinearB's 2026 mid year benchmarks, published August 4, 2026 on 2.7 million pull requests from 83,000 developers, put it in tiers. Pull requests overall merge within 30 days 90% of the time at elite organizations, 86% at good ones, 81% at fair ones. Agent opened pull requests merge at 79%, 58% and 37%. In the full year sample of 8.1 million, AI generated pull requests were accepted at 32.7% against 84.4% for human written ones, and waited 4.6 times longer before anyone looked.
Even at the elite tier, agents author at most 4.7% of pull requests. The factory, measured, is one pull request in twenty.
Mike Julian (@mikejulian), CEO of The Duckbill Group, wrote on X on September 6, 2026 that his team of five found itself with 60 open pull requests and two days of review ahead, and mostly ditched code review.
Inside our own line, the pull request that merges is the one that arrives with its tests, its evidence and a named engineer who approves it. The one nobody owns waits, then dies.
Sources
- LinearB, Your software factory needs a context layer (2026 mid year AI benchmarks), August 4, 2026. https://linearb.io/blog/does-your-software-factory-work
- LinearB, 2026 Software Engineering Benchmarks Report, public summary page. https://linearb.io/resources/software-engineering-benchmarks-report
- Mike Julian on X, September 6, 2026. https://x.com/mikejulian/status/2096665515708301424