If you use agents every day to create features, fix bugs, or diagnose problems, you need a solid workflow. With one in place, you should not have to keep cleaning up poor AI-generated code after every task.

We use a strong internal workflow set up by Stas Kokshin. You can also use Matt Pocock's workflow.

My usual workflow

  1. Grill the idea with the relevant documentation.
  2. Turn the result into a spec.
  3. Break the spec into issues.
  4. Implement the issues. This step already uses code review by default.

When the task needs an architecture decision or deeper diagnosis, add:

  • Improve codebase architecture
  • Diagnosing bugs

Other useful skills include:

  • Codebase design
  • Domain modeling
  • Test-driven development

Pstack by Lauren Tan is also worth exploring, especially if you use Cursor.

References

Originally posted on LinkedIn.