Skip to content
Page 6 of 6

Reflection 2

How This Reflection Works

Your Facilitator will pull from these questions to lead a cohort-wide discussion. If your AI assistant is building during a Challenge and you have a free moment, these also make great team conversation starters while you wait.

What You Built

  • What skills did your team create? Walk through one: what problem did it solve, and how did it change the consistency of AI's output?
  • You built your skill using the "We Do, You Do" pattern: do the process with AI first, refine until the output is right, then capture it as a skill. Why does doing the process first produce a better skill than trying to write the instructions from scratch? What would you miss if you skipped the "We Do" step?
  • What's the difference between asking AI to do something from scratch every time versus giving it a skill that encodes how you want the work done? How would you expect the output quality and consistency to differ?
  • When building iteratively with AI, each new feature can potentially break something that was already working. Why does this happen, and what can you do about it when your only verification method is manual review?
  • Can AI check its own work? If you asked an AI tool to review whether a codebase was well-organized or whether anything needed cleanup, how much would you trust its assessment? What are AI good at spotting in its own output, and what might it miss?

What You Practiced

  • Lesson 2 introduced decomposition: breaking a project into independently shippable pieces with a managed backlog. How does that change the way a team decides what to build next, compared to working from a vague list of ideas? What does "independently shippable" actually buy you?
  • What's the difference between checking AI's output against explicit acceptance criteria versus giving it a "looks good" glance? What kinds of failures does criteria-based review catch that a quick visual check misses?
  • Lesson 2 described the "spinning loop": re-prompting AI in circles when you don't have clear criteria for what "done" looks like. What causes teams to fall into the loop? What's the exit strategy?

How You Worked

  • How did your team divide the work this time? Did the decomposed backlog change how you organized? Could different people take different stories, or did you still need to mob?
  • Skills are meant to encode team conventions so that every AI conversation follows the same patterns. But do shared skills actually keep a team aligned, or can team members still end up with different conventions despite having the same skills? What else would you need?

Looking Ahead

  • Manual review works, but it doesn't scale. You walked through acceptance criteria by hand for every feature. How many features do you have now, and did you re-check the earlier ones after adding new ones? The faster you ship without automated checks, the more likely a new change quietly breaks something you already verified. What would it take to make that verification automatic?