The experiment
After a model upgrade, I normally ask the AI to inspect the code in my projects. This time I pointed the lens at myself. I asked it to study patterns across our conversations and explain how I think, where that style helps, and where it quietly creates friction.
The result was not a personality horoscope. It noticed a repeatable pattern: I think in systems, prototype quickly, and connect ideas across AI, games, music, automation, and interactive experiences. The same strength can create a prototype graveyard when the thrill of a new system beats the slower work of productizing the last one.
Why this works
A conversation archive is a behavioral dataset. It contains the problems you choose, the assumptions you repeat, the moments you change direction, and the evidence that persuades you. An AI can summarize those patterns, but only if you demand observations tied to examples.
Run the exercise
- Choose a workspace or chat history with enough real decisions to reveal a pattern.
- Ask for observations, supporting evidence, benefits, and costs.
- Separate high-confidence patterns from guesses.
- Convert each useful observation into one small operating rule.
- Review the rules after thirty days and keep only the ones that changed behavior.
My resulting operating rules
- Productize one prototype before starting three more.
- Document the pipeline so another person can reproduce it.
- Ship a public version before polishing the private one forever.
- Turn working demos into reusable templates.
Important boundary
This is reflection, not diagnosis. Treat every observation as a hypothesis and keep the final judgment human.
Software, links, and setup
- ChatGPT or another assistant that can reference only the conversations you intentionally provide.
- A notes app for the final operating rules.
- A calendar reminder for the thirty-day review.
Prompt pack
Pattern review
Copy this: Analyze patterns across the conversations I have intentionally shared in this workspace. Identify my thinking style, strengths, blind spots, and areas for self-improvement. For every claim, cite at least two behavioral examples from the supplied material. Label uncertain inferences as hypotheses. Do not diagnose me. End with five small operating rules I can test for thirty days.
Turn insight into action
Convert the observations into a table with: pattern, benefit, hidden cost, trigger, replacement behavior, and one measurable thirty-day experiment. Prefer behaviors I can observe over abstract advice.
What to remember
- Ask the model for evidence, not flattering labels.
- Turn observations into experiments instead of identities.
- Keep the human in charge of interpretation.
- The useful output is a changed habit, not a clever analysis.
Then test it, document it, and make the next version easier for someone else to reproduce.