Coding with AI tools

Stage 10 of 10 · 5 lessons · ~4 weeks

A real project shipped with an agent

Put it all together: scope a small real project, write a spec, build it in vertical slices with your tool, gate every pull request on CI, disclose AI use honestly and publish a write-up of your decisions.

Start: Scoping a project you can finish

Lessons· one lesson for each topic, in this order

  1. 1Scoping a project you can finish
  2. 2A written spec and vertical slices
  3. 3CI and one pull request per slice
  4. 4Deploying and learning from real use
  5. 5Disclosing AI use and writing it up

Do this

  • Scope a project with a real user, three to five features and a written out-of-scope list.
  • Write SPEC.md with a done-when check for every feature.
  • Set up CI with tests, type checks and scanners before the second slice.
  • Ship each slice as its own small pull request with a clear description.
  • Deploy early and watch a real user try it.
  • Disclose AI use in the README and publish a write-up.

Start with· follow one of these

  • CourseFree
    Claude Code in Action

    Anthropic · Intermediate

    Free course lessons on longer sessions, GitHub Actions and verifying work you did not watch

  • GuideFree
    How to Write a Git Commit Message

    Chris Beams · Beginner

    Seven rules for commit messages that make your project history readable

Keep handy

Also worth your time

Before moving on: A small real project deployed and used by at least one real person, built in reviewed slices with CI on every pull request, with honest AI disclosure in the README and a published write-up of your decisions, the agent's role and the bugs you caught.

Tauseef's advice
This project is your proof that you can direct an agent without being directed by it. Keep it small, real and finished. The write-up matters as much as the code: it is where you show the judgment that no tool can fake.

Mark this stage as not done?