LinkedIn has built a multi-agent AI system that reviews more than 40,000 pull requests a week, ByteByteGo says. Engineers accept nearly 64% of its comments, offering a concrete example of AI being used to tackle the review bottleneck created by faster code production.
Watch Desk analysis
What happened
ByteByteGo describes an in-house system built around an orchestrator and subagents. It applies rules at company, repository and pull-request level, then generates, merges and checks its comments before they reach a person.
The reported acceptance rate is nearly 64%. ByteByteGo says comments about concurrency bugs and dead code had the highest acceptance rates. The system is intended to support human review, not replace it.
Why it matters
When code arrives faster, the scarce resource can become the attention needed to review it properly. LinkedIn’s approach points to one answer: use AI to sift through a large queue and flag issues, while leaving engineers to decide what deserves action.
The acceptance figure is useful because it goes beyond a polished demo. It suggests engineers find a substantial share of the comments worth keeping. But it is not a measure of how many bugs the system catches, or proof that the comments it misses are harmless. Those are different tests, and the distinction matters when code moves from suggestion to production.
Our read
This is a practical use of AI with a clear job and a human still holding the pen. The layered rules and pre-review checks sound more like an attempt to fit the tool into real engineering work than a bot given the keys and wished good luck. The strongest next evidence would be how the system affects review time and the quality of code that ships, not just how many comments get accepted.
What to watch
- Whether LinkedIn shares results on review speed or defects caught.
- How engineers handle comments the system generates but reviewers do not accept.
- Whether the approach is adopted beyond LinkedIn.
Discussion spark: For AI code review, which should teams value more: the share of comments engineers accept, or evidence that the system catches important bugs without slowing releases?
Sources and evidence
- How LinkedIn Uses AI to Review 40,000 Pull Requests a Week (9 November 2026, 16:30 UTC)
- At Northern Trust, finance experts and Microsoft engineers partner to move faster on AI – Microsoft Source (9 November 2026, 16:30 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.