Writing
Where AI coding agents break on real Apple work.
Short, specific write-ups on how AI coding agents perform on real Swift and Xcode work, and the failure modes that show up once the code already compiles. A new post most Mondays.
Where AI agents break on Swift
- Watch a coding agent silence a Swift 6 data race instead of fixing it Hand a coding agent a Swift 6 strict-concurrency error and it often makes the build green by silencing the check, leaving the data race in place.
- Coding agents are good at writing Swift. They're bad at finishing it. Frontier models write good Swift. Where coding agents come apart is finishing the work: intent mismatches, races that compile, and fixes that loop.