The future of AI coaching

Episode Summary

Executive Summary: Stanford’s James Landay argues that LLMs can do more than generate text: they can act as personalized coaches and tutors. He describes GPT Coach for fitness, adaptive to user barriers and health data, and Smart Primer/ACORN for outdoor, story-based elementary learning that boosts writing and knowledge. He emphasizes human-centered AI that considers users, communities, and society.

Main Topics: LLMs as Fitness Coaches (Priority: 5/5): Landay explains how GPT Coach uses large language models, motivational interviewing, and wearable/phone data to create personalized fitness plans that account for real-life constraints. Interdisciplinary Design and Health Expertise (Priority: 4/5): The coaching system is built with input from Stanford health/public health experts, including a structured manual and consultation to shape coaching style and content. Ambient Awareness and Display Design (Priority: 4/5): Landay revisits his research on ambient displays—subtle visual cues on phones/watches—to keep users aware of progress without requiring explicit app use. LLMs for Elementary Education (Priority: 5/5): The Smart Primer and Moon Story projects use narrative, augmented reality, and LLMs to teach science and sustainability outdoors, making learning more engaging and personalized. Encouraging Writing Through AI Feedback (Priority: 5/5): In the Moon Story project, the LLM helped children write more by asking follow-up questions and providing personalized responses, countering fears that LLMs will destroy writing. Human-Centered AI (Priority: 5/5): Landay argues AI design must go beyond user-centered design to include community and societal impacts, since AI often affects people beyond the direct user.

Key Arguments: AI can be effective as a coach when it handles qualitative barriers, not just quantitative goals. Personal coaching is valuable but too expensive and scarce for most people; LLMs can help scale it. Motivational interviewing is a useful framework for keeping AI coaching grounded and supportive. Wearable and app data can enrich coaching by providing context over months, not just during one interaction. For education, narrative and outdoor, embodied activities can engage children who do not thrive in conventional classroom settings. LLMs can increase, not decrease, student writing when they provide personalized feedback and follow-up prompts. Human-centered AI must account for downstream effects on non-users, communities, and society, not just the person operating the system.

Data Points: Project timeline: 10- or 20-year efforts - Landay describes education and health as long-horizon research problems. Fitness data window: 3 months - GPT Coach can ingest three months of prior wearable/phone data to inform coaching. Prototype meeting length: 30 to 60 minutes - Current coaching prototype replicates an initial coach-intake session. Short-term study length: 3 weeks - Planned short study to debug the coaching interface before longer trials. Long-term study length: 3, 4, or 6 months - Desired study duration to test behavior change and goal attainment over time. Project history: since 1995 - Landay credits Neil Stevenson’s The Diamond Age as the origin of his idea for the educational system. Prior work milestone: 2010 - He says the iPad made the story-based, mobile learning concept seem feasible.

Pivotal Quotes: "AI is not just good at creating text and answering our questions. It can motivate us as a coach, and it can teach us as a tutor." — Russ Altman: Episode framing statement introducing the conversation with James Landay. "The LLM allowed us to tell what they had written, and sometimes you get a kid who just writes almost nothing. And we were able to use the LLM to encourage those kids who didn't write much to actually write more." — James Landay: Explaining the Moon Story learning result where AI increased student writing. "Human-centered AI means we need to design at the user level, but also community level and society level and think about all those together when we're designing AI systems if we want them to have a positive impact." — James Landay: Landay defines the broader design philosophy behind his institute’s work.

Implications: LLMs may become practical coaches and tutors, not just chatbots, especially when paired with design, domain expertise, and real-world context. The biggest opportunity is personalized support that scales without losing human-centered safeguards.