antirez: RL will push AI-generated code further from what humans consider good
antirez · x · 2026-09-27
Continuing the discussion with mitsuhiko, antirez argues that due to RL effects, the way AI writes code will diverge more and more from what we used to consider good and comfortable. He now only evaluates outcomes for extended periods — as long as the code delivers and the system understands his big-picture ideas, he's fine. He also laments having no time yet to hand-write small poetry-style programs.
More from coding & agent
- TypeSafe ships official skill teaching Claude Code to batch Jev calls, cutting bills 12.2x — anselm · 2026-09-27
- World's fastest panel QR factorization on B200: how a GPU MODE contestant cracked chained dependencies — A_K_Nain · 2026-09-27
- Graph Engineering: Author Finds Agent's Self-Checking Was Lying to Him — blaizedsouza · 2026-09-27
- One prompt in OpenCode builds a full 3D solar system observatory with themes — JaynitMakwana · 2026-09-27
- Stop building agents as a straight line: the 90-second test to cut fake dependencies — blaizedsouza · 2026-09-27
- Digital Apprentice framework: agents earn autonomy through per-skill authorization gates — blaizedsouza · 2026-09-27