Why AFitPilot® exists

We set out to make AI replace the coach. A year of trying taught us the opposite.

Team

Currently building AFitPilot.

Walter Clayton

Founder

I spent a decade coaching athletes before writing a line of the code that became AFitPilot — and that order matters. Most training apps are spreadsheets with a login. They hand you a plan and hope it fits. Coaching is the opposite: you watch how someone responds and you adjust. That's what I've spent years trying to teach software to do. AFitPilot is the result — adaptive programming that reads training the way a good coach would, built by someone who's stood on both sides of the whiteboard.

Connect on LinkedIn

Vito Krizmanić

Sports Science & Product

Kinesiology student at the Faculty of Kinesiology, University of Zagreb, finishing his final exams before going full-time on AFitPilot. A former elite handball player with 14 years of competitive play, plus 10 years of weightlifting and fitness coaching — combining the mindset of an athlete with the analytical approach of a sports scientist. At AFitPilot, Vito works across sports science, product, marketing, and code — bridging the research and the day-to-day of what coaches actually need.

Connect on LinkedIn

How we got here

From a GP's home computer in 1971 to a coaching platform in 2026.

1971
The origin
AFitPilot didn't start in 2019. It started in 1971, in a GP practice in Norwich, with a man who taught himself to code at evening classes. Dr. Geoffrey Clayton — GP, pioneer, and my grandfather — built Takeheart, an expert system that turned patient data into personalised health reports. The intellectual architecture behind AFitPilot is the same.
2019
Where I started
Six months into learning to code, I built a Heath-Carter somatotype calculator. Vanilla JavaScript, a hand-drawn canvas, the academic references I'd kept from university. The vision was already there: take measurements, compute something specific, return something meaningful.
2022
A real application
Three years later, I rebuilt the somatotype as a full MERN-stack web app — React and TypeScript on the front, Express and MongoDB on the back, with accounts, a dashboard, and a blog. The first time the AFitPilot idea looked like a product.
Today
Today
AFitPilot 2.0 — built on what didn't work
A year ago I was building programs in ChatGPT — PDFs, Google Forms, me overriding the model every session. It worked because I was in the loop. In December 2025 I tried the opposite: athletes log, the LLM generates next week. Without a coach, the model hallucinated and athletes didn't engage. The lesson: the coach stays in the loop. AI removes the grunt work; it doesn't replace the judgment. AFitPilot 2.0 launched in May 2026 on that insight.
- Two-stage AI generation — strategy first, then exercise selection from a filtered library
- Plan memory — context events, coach overrides, and tendency profiles persist across weeks
- Coach override at every level — exercise, block, session, full plan, with hard bans on exercises, patterns, or equipment
- Foster sRPE × duration load tracking, EWMA-smoothed trends, Hooper-Mackinnon daily readiness
- Closed-loop quality — every coach correction logged as a before/after event and replayed against new code before it ships