danluu: 'Brain-off' LLM coding works better than ever — and still ends badly for the programmer
threepointone · x · 2026-09-19
Engineer danluu argues that ever more people are using LLMs with their brains switched off: accepting generated code at face value and looping errors back to the model, a pattern he calls being a "meat proxy" for a for loop. In early 2025 this mostly failed; by September 2026 he finds the resulting software surprisingly, if imperfectly, functional. But his core point is a career logic: if brain-off development ever produces good software, the company has no reason to employ the meat proxy — it can just run the LLM in a loop. So the methodology fails for the employee at every stage, either because the model isn't good enough or because it is. With thanks to Yossi Kreinin, Thomas Dullien and other engineers for discussion.
More from AGI Musings
- Developer laments serious coding skill atrophy after months of relying on AI — ShikharMurty · 2026-09-19
- Nick Bostrom admits AI panic 'feels out of control,' gets mocked as its author — NinaDSchick · 2026-09-19
- RL isn't a cherry on the cake: pretraining is just initialization for RL — maksym_andr · 2026-09-19
- Ex-Apple engineer recalls China's 996 workers napping 2 hours at desk, less productive than UK 9-to-5 — alexvoica · 2026-09-19
- Opinion: Normalize Treating Frontier AI Community as Fringe Activists — norcalnatv · 2026-09-19
- Seroter's reading list: Zed's Delta kills PRs, agents pick the infra — rseroter · 2026-09-19