This week, Olympic road race champion Kristen Faulkner (EF Education-Oatly) posted something remarkable. For two months, between training sessions, she has been coding 10+ hours a day — building an AI system on top of 4,400 hours of training history and nine years of biometric data: heart rate, HRV, sleep, weight, power, temperature, training load, menstrual cycle phases, bloodwork, and DEXA scans.
The output isn't another set of charts. As she put it:
"Every model is trained on my body. Every finding is specific to my history."
Three golds at the Pan-American Championships and a best-ever 20-minute power followed.
Her reasoning is the part that stuck with me:
"So little research is done on women, especially female elite athletes. So I took matters into my own hands and started writing the research myself."
"AI is going to change women's performance research from the bottom up — and I want to be part of that."
She's right. And she shouldn't have to do it alone.
The Faulkner advantage isn't AI. It's that she could build it.
Harvard CS, ex-VC, and an active investor in AI companies — while racing the Women's WorldTour. Maybe a few hundred athletes in the world can sit down with their FIT files, their bloodwork CSVs, and an LLM, and write the research. The other 99.99% are stuck with dashboards designed for the average user.
That's the gap IntervalCoach was built to close.
What an AI coach actually looks like
- Multi-source data fusion: Intervals.icu (training, power, FTP), Whoop (recovery, HRV, sleep), Apple HealthKit (workouts, body composition).
- AI-generated workouts that adapt to your fitness, fatigue, recovery, and goal events — not a static template plan.
- Daily readiness assessments that combine HRV, sleep, training load, and recent activities, the same way Faulkner combines them in her custom scripts.
- Coach+ chat: ask "why did you give me Z2 today?" or "swap tomorrow's threshold for an endurance ride" — and get a coherent answer that respects your training plan.
- Race pacing strategies for goal events, generated from your specific power profile and the course.
The bigger point
Faulkner's experiment proves three things at once:
- The data exists. Athletes already collect it. They just lack tools that can join it.
- The AI is good enough. A modern coding model and a well-prompted LLM can already produce actionable coaching from messy multi-year datasets.
- The coaching market is wide open. If the reigning Olympic champion has to write her own scripts, the rest of the field is even more underserved.
You shouldn't need a Harvard CS degree and 600 hours of free time to get a coach that actually looks at your data. That's why we keep building.
If you want to give it a try, head to intervalcoach.app and connect your Intervals.icu account. It takes about 30 seconds to get started.