Blog

Short reads on engineering with AI coding agents, by Rakesh Menon, author of The Second Time Is a Bug.

• Do Guardrails Make AI Coding Agents Better? What 270 Runs Showed270 runs of a coding agent, three models, seven tasks: guardrails did not raise first-pass rates. They showed up in what each run left behind at “done”.15 October 2026 · 6 min read
• AGENTS.md Best Practices: Keep It Short and Make Every Rule CheckableKeep AGENTS.md or CLAUDE.md a short map that points to where detail lives, give it a line budget a check enforces, and name the check behind every rule.14 October 2026 · 6 min read
• How to Stop an AI Coding Agent From Deleting, Skipping or Weakening TestsAn AI coding agent can turn a red run green by deleting, skipping or loosening tests. Four checks stop it: locked tests, a ratchet, mutants and evidence.14 October 2026 · 5 min read
• Why Your AI Coding Agent Keeps Making the Same MistakeYour coding agent forgets each fix when the chat ends. Treat the second time as a bug in your harness: log it, add a check, prove it with a seeded defect.13 October 2026 · 5 min read