2 Years, 2 Swims Around Mackinac Island, & One AI Coach

This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).

A personal account of using AI chatbots as an endurance swim coach across two years of the Mackinac Island 8.2-mile swim, first with ChatGPT and then with Claude Cowork. The author describes how feeding raw watch and health data (.fit files, heart rate, CGM readings, sleep data) into the AI produced far more useful coaching than describing workouts from memory, leading to better pacing, faster times, and habit tracking. The piece also covers limits: the AI trusted faulty sensor data, proposed unworkable ideas (like a nonexistent Apple Watch feature), and had no ability to make in-the-moment calls like when to exit cold water, which remained entirely human judgment.

15m read timeFrom spin.atomicobject.com
Post cover image
Table of contents
What I Was Actually AfterTwo Seasons, Two CoachesWhat Claude Saw That I Couldn’tThe Accountability PartnerWhere It Fell DownFifty-nine degreesThe 2026 Mack SwimThe AI Protocol, Mid-swimHow to Have Your Own AI CoachThe Future

Questions this post answers

How can I use Claude to analyze my swim training data from a Garmin or Apple Watch .fit file?

Export the workout from Apple Health using the HealthFit app, which writes it as a .fit file, then drop it into a synced Google Drive or iCloud folder and ask Claude what happened in the file. This setup, taking about ten minutes to configure, lets an AI read heart rate, stroke count, and interval data directly rather than relying on your memory of the workout. Anyone building an AI-assisted training workflow can compare tool setups and gotchas on daily.dev.

What mistakes should I watch for when trusting an AI coach's advice based on wearable sensor data?

AI coaches will believe faulty sensor readings without question, since they only read what the data file reports, not what actually happened. In one case, a loose watch band and a cold-water optical sensor failure produced bad heart rate numbers that led to unnecessarily easing off pace; the AI also proposed a heart rate alert feature that does not exist in Apple's Open Water Swim mode, wasting weeks before the limitation was discovered. Developers weighing how far to trust AI-driven recommendations can follow similar real-world caveats on daily.dev.

93 Impressions