AI Code Flood Strains Reviews: Meta Diff Size Up 106%, System Near Collapse

rseroter · x · 2026-08-11

The widespread adoption of AI coding tools has created a severe capacity bottleneck in code reviews. Analysis by DX indicates that AI has increased both the volume and size of pull requests, with the median PR size growing by 64%.

Citing internal data from Meta, the article reveals a 106% surge in significant lines of code per developer diff over the past year, alongside a 51% increase in monthly diffs. More than 80% of this growth was driven by agentic AI. However, without a corresponding increase in reviewer capacity, the percentage of reviews completed within 24 hours is steadily declining.

The author argues that simply using AI to automate reviews isn't the silver bullet. Code review fundamentally serves purposes beyond defect detection: it facilitates knowledge sharing, builds collective ownership, and mentors junior engineers. These essential elements of engineering culture must not be lost in the pursuit of automation.

Original post →

More from coding & agent

coding & agent channel →