Developer's Guide to Coding with AI

  1. Treat AI as a tool of thought, not a replacement for thinking.

    It is an extension of your editor, not your engineering judgment.

  2. It is a multiplier.

    Strong engineers get sharper with agents. Weaker engineers cause more damage. There is no shortcut around understanding the problem.

  3. Start small, raise expectations gradually.

    Begin with boilerplate, merge conflicts, and repetitive refactors. When those go well, push toward harder tasks.

  4. Always validate.

    More tests, more ways of testing, more fuzzing, more randomization. The headroom in agent-assisted work is in your CI, not in the prompt.

  5. Use the latest models, and keep at least two providers handy.

    Model providers experience downtime, sometimes daily. Switch between Claude Code, Codex CLI, and others.

  6. Save guidance to CLAUDE.md or AGENTS.md, but keep it short.

    Long instruction files get ignored. Avoid telling the model what not to do, as that often has the opposite effect of what you intended.

  7. Be specific.

    Agents reward complete specifications. Saying exactly which files, which functions, and which approach gets better results than vague prompts, and it preserves your engineering skill in the process.