Code like a pro with AI: plan, tests and review

What you'll see in this lesson

1. The workflow that prevents fragile code

   The teams that work best with agents always repeat the same cycle: plan, tests, code, review and commit.

2. Plan mode: first it reads and proposes, then it touches

   In plan mode the agent explores the code and presents you a plan, but changes nothing until you approve it.

3. Big tasks: interview and specification

   Before a big feature, ask the agent to interview you and put everything in writing in a file.

4. Tests first: the agent knows when it's done

   With TDD (test-driven development) you write the tests first, check that they fail, and only then write the code.

5. Give the agent something it can check

   Anything that returns «pass» or «fail» works: tests, build, linter or a screenshot compared with the design.

6. Several sessions in parallel with git worktrees

   A worktree is another working copy of the same repository, in its own folder and branch: each agent works in its own.

7. Automatic review before accepting each change

   A reviewer with a clean context finds bugs the agent that wrote the code can't see.

8. Small commits: every step, a way back

   Commit every time something works and passes the tests: one change, one commit, one clear message.

9. Summary: the professional workflow with agents

   Plan, tests, verification, parallel work, review and small commits: slower at first, much faster in the end.