Matt Pocock: The Scariest Part of AI Coding Is Not Feeling Your Strategic Mistakes
mattpocockuk · x · 2026-09-10
Matt Pocock argues the hidden danger of AI coding is that you lose the felt pain of strategic mistakes—like his 2018 choices of an overbuilt form library and a dumb webpack+PHP combo, which the team didn't even recognize as mistakes at the time. Agents, he notes, are trained to execute the task given, not to propose strategic alternatives, and humans can't rely on pain to spot them. He's building a /retro skill to let agents surface refactor opportunities, though he remains nervous about it.
More from coding & agent
- Users Petition OpenAI for $400-$600 Heavy Builder Tier as $200 Plan Runs Dry in 48 Hours — dragonwarrior_1 · 2026-09-10
- Codex founding engineer explains why Rust, why open source, and how the harness works — mathemagie · 2026-09-10
- He used Grok to fix X's broken archive viewer and merge posts into one AI-searchable JSON — davidpattersonx · 2026-09-10
- banteg crowdsources AI agents to finish matching Crimsonland's original x86 code — banteg · 2026-09-10
- The agent loop is the seventh draft: tracing agentic execution from Windows 3.1 message pumps — bibryam · 2026-09-10
- A visible screenshot doesn't mean the model got it: a 3-step agent image-path diagnostic — derspenti · 2026-09-10