Skip to content

Reflection 3

A facilitation guide for the team debrief following Run 3. These questions are designed to spark discussion — not every question needs to be covered. Pick the ones that resonate with what you observed during the run.

What You Built

  • How many automated tests does your project have now? Did any of them catch a regression — a feature that broke when you added something new? What would have happened without that test?
  • What's the most ambitious feature you added during this run? Would you have attempted it without the safety net of automated tests backing you up?
  • Is there a feature on your live URL right now that you're genuinely proud of? Pull it up — what makes it work?
  • Is your application getting harder to change? Are new features taking longer than early ones, or are changes causing unexpected side effects? If so, that's a signal that the codebase may need reorganizing before you add more to it.

What You Practiced

  • Did the closed loop — criteria → test → fail → implement → pass — change the way you delegate to AI? How is handing AI a story with tests already written different from handing it a story and checking manually after? Did you notice AI responding differently when it had a failing test to work toward versus just your description of what to build?
  • Did you create a TDD skill or update your project context file to make the closed loop automatic? If so, did it work — did AI follow the red → green pattern without you spelling it out each time?
  • Lift 3 talked about the two-week cliff — changes silently breaking verified work. Did you experience anything like that during this run, and did your tests catch it? Or did the safety net hold?

How You Worked

  • With automated tests handling verification, did your team's workflow change? Could you work faster, take bigger swings, or divide the work differently compared to Run 2?
  • Think about accountability. In Run 1 and 2, you were responsible for AI's output but could only spot-check it. Now you have tests. Did that change how comfortable you are putting your name on what shipped? What's the difference between "I looked at it and it seemed fine" and "the tests pass"?
  • Did anyone on your team try running the live URL on their phone or sharing it with someone outside the team? What was the reaction?

Looking Ahead

  • You're building one feature at a time — write criteria, generate tests, implement, verify, ship. It works, but you're still the bottleneck. What would change if you could delegate multiple stories in parallel and have AI work on them simultaneously?