Modern coding is building good evals: 'your code is bad probably because the verifier is bad'
aidenybai · x · 2026-10-13
Aiden Bai argues modern coding is essentially building good evals: with proper verifiers and environments — lints, code review agents, processes, tests — you can 'trust' agents to write correct code. His sharp corollary: if your code is bad, it's probably because the verifier is bad, or the task (e.g. design) is just hard to verify objectively.
Related event: Aiden Bai: Modern Programming Is About Building Evals(2 posts)→
More from coding & agent
- LangChain Interrupt London: Harrison Chase keynotes on owning your intelligence — marlene_zw · 2026-10-13
- Why MMR keeps returning duplicate chunks to your agent: check candidates_limit first — izgorodin · 2026-10-13
- Survey: Evaluating LLM Outputs Is the Top Bottleneck for 49% of NLP Builders — _inesmontani · 2026-10-13
- UX Without a UI: Designing Terminal Experiences for an Agent-First Product — dl_weekly · 2026-10-13
- Stalkr adds AI translation to flag and translate foreign-language brand mentions — marclou · 2026-10-13
- LangChain Interrupt: visual agents at DB and mission-critical agent deployments — LangChain · 2026-10-13