Setting up a codebase for working with coding agents (simonwillison.net)

A developer on Hacker News laid out practical tips for making a codebase productive for "coding agents" like Claude Code. The essentials: a good automated test suite (they use pytest — one project has 1,500 tests) so the agent can run only the tests relevant to a change and then the whole suite at the end; the ability for agents to interactively test code by starting a dev server and using Playwright or curl; and maintaining a GitHub-issues collection with direct URLs that agents can consult. Also recommended are linters, type checkers and auto-formatters so agents can run the same tooling humans do. The author notes agents can read code quickly, so extensive docs are less critical for agent use, though agents help spot stale documentation. For the AI/ML community this is significant because it shifts best practices from purely human ergonomics to agent-first workflows: robust tests, reproducible dev environments, and machine-readable issue/context links materially improve an agent’s reliability and speed. Technical implications include prioritizing selective test execution, CI that supports incremental checks plus full-suite validation, and exposing local endpoints for automated interaction testing. In short, anything that makes code easier for humans to maintain tends to make it more usable for LLM-based coding tools — enabling safer, faster, and more automated development loops.