217: "How Do You Create AI Coaching That Adapts Over Time?" (Reflections On Gavin Lorenzo)
🧠 Erik’s Take
Erik treats this “short review” as a real question, not a tech demo. The core shift for him is moving from AI personalization as memory, to AI personalization as understanding: a predictive model of how a specific person learns, gets stuck, adapts, and responds. He also feels the conversation complicate the regulation instinct. It is not just “should we regulate,” but “who gets to decide what these tools can do” and what values get embedded. Finally, he leaves with a more intellectual spark: if AI can model people well enough, it could turn theories of human behavior into operating hypotheses,
🎯 Top Insights from the Interview
- Understanding is not the same as remembering. Erik reframes Tellbloom-style personalization as building a predictive model of how someone operates over time.
- Regulation is a second-order question of governance. The real issue is who decides access, capabilities, and embedded values.
- AI-enabled human modeling could enable falsifiable learning loops. Erik highlights the move from “a theory sounds right” to “does it improve the interaction.”
- Ethics is not cleanup work after launch. Erik emphasizes privacy, consent, and transparency as product design constraints, not paperwork.
🧩 The Personal Layer
Erik notices his own friction point: he agrees AI needs guardrails as capability grows, yet he worries regulation can be captured, slow, or incentivized toward incumbents and politics. He also recognizes the temptation to reduce personalization to “better memory” because it is measurable and easier to ship. The episode pushes him to hold a more demanding standard: if the interaction is coaching, education, leadership, or decision support, then the product has to adapt with real human-level nuance, not generic helpfulness.
🧰 From Insight to Action
- If you are building or buying AI that interacts with people, define “understanding” as a behavior-improving capability, not just personalization as recall.
- Design governance into the product. Make transparency, retention limits, and consent visible to users from day one.
- Treat people frameworks as testable operating hypotheses. Instrument the experience so you can measure whether a model improves outcomes for that person.
- When discussing regulation, shift the conversation from slogans to decision rights: who sets capabilities, who audits, and how values are encoded.
🗣️ Notable Quotes
- “Understanding isn’t the same thing as memory.”
- “The deeper question is: who gets to decide what these tools can do.”
- “Move from ‘this theory sounds right’ to ‘this model either improves the interaction or it doesn’t.’”
- “If your product helps AI understand people more deeply, then privacy, consent, transparency, and use aren’t cleanup work later.”
🔗 Links & Resources
