The AI Worked, but the Learning Didn't! with Sarah Wood

AI can simulate a conversation, we all know LLMs are very good at sounding conversational. That's table stakes. Now the interesting part is what happens when that conversation isn't grounded in real work, doesn't represent realistic resistance, hasn't been validated by a human, and gives feedback nobody can connect to performance. This session is a failure analysis of a deliberately flawed AI-supported roleplay. We’ll look at a scenario where the AI technically “works”: it responds, adapts, scores, and gives feedback, but the learning design underneath is weak. Participants will use a four-part workflow model from my own enablement practice: Contextualize, Co-create, Validate, Measure. Contextualize: what real work is this tied to, and what behavior should change? Co-create: what should AI help produce, and what design decisions still belong to us? Validate: who needs to test whether this is accurate, safe, realistic and useful? Measure: what behavior signal would tell us whether anything changed beyond completion? Attendees will diagnose the failure, redesign a single element, and stress-test whether that change makes the experience more grounded, realistic, validated or measurable. This is about the uncomfortable bit: AI can make weak learning design look convincing, and that makes our design decisions matter more than ever.