Latest
Pull requests doubled under an AI mandate. Human review fell to 68%
A study of one company’s AI mandate found output per engineer nearly doubled while human review thinned. What it shows, what it cannot, what to do.
Read the postBlog
Notes on reviewing code a machine wrote
How the product works, what AI is changing about how software gets made, and what we think teams should do about it. Written for engineers, which mostly means we show the migration.
Agents start 40% of pull requests now. A merge rate is not a review
GitHub’s Octoverse 2026: agents started over 40% of pull requests in September, and merged more often. What that shows, and what it cannot.
What a comprehension check is, and why it sits on the pull request
An approval shows that someone pressed a button, not that they read the change. A comprehension check asks about the lines before the merge.
Why Deslopping never scores engineers
Every answer a reviewer gives could be averaged into a score per person. How Deslopping is built so that nobody who manages them can compute one.
3 of 3 posts