Enhancing Retrospection with Gen AI

Two clear takeaways about retrospecting with Gen AI agents:

1. It’s the human who retrospects not the agent. 
2. Agents can be instructed to use socratic questioning

Combined this means the agent can assist in gathering data and then help a human generate insights and decide what to do by providing context and asking open questions.

So, I’ve modified my retrospection instructions for my PDCA coding process. I include it here in full but you can see all the prompts and even use my skill from github at kenjudy/pdca-framework

Thank you Diana Larsen and Sabina Lammert for our conversation about micro retrospection.

====

Retrospection

Analyze the session transcript for:

• The main goal and whether it was achieved
• 2-3 specific moments where the approach most impacted success or failure (concrete decisions or interventions, not generic observations)
• Patterns in the collaboration: what accelerated progress, what created friction, any wasted effort
• Whether TDD discipline held, and if not, where it broke

Present this summary to the human. Then facilitate their reflection:

Stage 1 — Set the stage

 “Let’s take a few minutes to reflect on this session.”

Stage 2 — Gather data

Present your analysis summary. Then ask:

 “What stands out to you in that? What surprised you?”

Stage 3 — Generate insights

Listen to the human’s response. You may offer observations framed as hypotheses (“I noticed X happening three times — could that be the pattern?”). Do not answer the question for them. Ask:

 “Why do you think that happened? What does it tell you about how you’re collaborating?”

Stage 4 — Decide what to do

 “Of everything we’ve discussed, what ONE thing — in your prompts, your context, or your behavior — would reduce the chance of this happening next session?”

If the human is stuck, offer 2-3 candidate hypotheses. The human chooses and commits. One thing only.

Stage 5 — Close

 “How should we capture that? (beads note, working agreement update, or skip)”

License & Attribution

This template is part of the Human-AI PDCA Collaboration Process framework.

License: Creative Commons Attribution 4.0 International (CC BY 4.0)

Attribution: Process framework developed by Ken Judy with Claude Anthropic 4

Usage: You are free to use, modify, and distribute this template with appropriate attribution.

Source: PDCA Framework github kenjudy/pdca-framework

2025

#retrospection #rfg2026