Matt Pocock: In the Agent Era, Code 'Readability' Is a Nicety — Optimize for Verifiability Instead
mattpocockuk · x · 2026-11-04
Matt Pocock argues that agents change what matters about code quality. Humans read text slowly, so readability used to be essential; agents 'read' by executing and poking at code to gather data, so what actually matters is verifiability — how easy it is to experiment with code to confirm it works.
Key points:
- A huge module full of performance tricks with 100% test coverage takes a talented human an hour or two to get comfortable with; an agent needs zero time — it verifies via tests and explores.
- Readability now only matters for two things: humans can jump in for spot checks on important PRs (observability), and code should share the domain language humans use to describe it (helps agents translate requests into changes).
Takeaways: judge modules by how easy they are to verify, not read; design codebases so modules can be tested independently; feel free to trade some readability for performance — cyclomatic complexity and function size don't matter to agents.
More from coding & agent
- Postman desktop app reported to auto-install Claude Code/Codex plugins; opt-in to be clarified — shoumikgoswami · 2026-11-04
- Opus's viral takedown of the 'vitrine-chic' AI web demo style: built for the screenshot, not the session — teortaxesTex · 2026-11-04
- 3 MCP servers ate 35k tokens of context — dynamic tool loading fixed it — Jaig5970 · 2026-11-04
- Dev shares 8 identity-mapping rules after his WhatsApp agent matched the wrong person 3 times — petrusenko_max · 2026-11-04
- DNS inventor Paul Mockapetris and Vint Cerf debate DNS as a discovery layer for AI agents — IvyTatiana88 · 2026-11-04
- OpenAI acquires Pierre Computer to bring code storage into Codex, challenging GitHub — kimmonismus · 2026-11-04