Good LLM app operators follow Toyota's genchi genbutsu: observe real outcomes
charles_irl · x · 2026-09-18
charlesirl applies Toyota's genchi genbutsu principle to LLM engineering: good factory operators observe real outcomes in their actual context, and the same held for classical ML pipelines — Karpathy learned to solve ImageNet classification himself. Running LLM applications is no different.
In a reply he adds that not looking at the code violates the San-Gen Shugi principle and won't be part of successful software factory management.
More from coding & agent
- Developers question why Anthropic still hasn't adopted the agents.md standard — Rasmic · 2026-09-18
- Replace agent classification steps with optimized small models to cut latency and cost — LangChain · 2026-09-18
- InfoWorld: Software developers are all managers now in the age of coding agents — rseroter · 2026-09-18
- 15 token-saving tricks for Claude: cut ~19,700 tokens per fix by scoping rewrites — goyalshaliniuk · 2026-09-18
- Vibe coding produces 50K-line PRs, but nobody can verify they actually work — srchvrs · 2026-09-18
- AutoData: EMNLP paper uses agentic search to automate pre-training data selection — ChengleiSi · 2026-09-18