Press Space to continue
Evaluation becomes the bottleneck as code generation gets cheaper
Press Space to continue
When code gets cheap, the bottleneck moves to evaluation. Not prompt engineering — everyone says that. Actual evaluation. Knowing which of the fifty functions your agent wrote is correct, and having the harness to verify it. That's where the hard work is in 2026.
open on x · 1 viewViews and clicks count once per visitor per day. Clicks are opens of this post on X.