The research did not exist for her body
Kristen Faulkner, the Olympic road race champion, went looking for sport science that fit her and found little. "So little research is done on women, especially female elite athletes," she told Cyclingflash. So she built her own system: a personal AI trained on nine years of her biometric data, including heart rate, heart rate variability, sleep, weight, power, menstrual cycle phases and body scans.
Specific input, specific output
Her description of the result: "Every model is trained on my body. Every finding is specific to my history. And every output is actionable, not just interesting." She reports her best 20-minute power ever "with the help of this app."
What the evidence does not show
Be exact about the claim. None of the reports separates the AI's contribution from her training, racing and everything else, and she is not reported claiming the AI did it alone. The wattage was not published. What can be said is narrower and more useful: the thing she changed was the input.
Why general questions return average answers
A general-purpose tool answering a general question has only the general case to draw on, so it returns the answer that fits most people. That is useful for a first orientation and close to useless for a decision that depends on your own pattern. The records are what turn the second kind of question into one the tool can actually work on.
The portable part
Most people ask AI about their work in general terms and receive the average answer back. Faulkner gave it years of her own records and asked questions only that history could answer. The difference between a generic answer and a specific one was the data she supplied.
A team's version is already on disk: a year of proposals, tickets, decisions and handoffs that no general-purpose tool has seen. Pick one question that industry research cannot answer for you and put your own records to it.
Starting does not need nine years or a purpose-built system. It needs one body of your own material, a question that depends on it, and a way to check the answer against what you already know about your own work.
Practice the question before it matters: Practice putting a real question to your own records →
Keep reading: In AI, Context is King · Two questions about any idea. You can only answer one of them yourself.
This is the thinking behind a Trusted AI Culture.
Sources: Cyclingnews, 23 April 2026; Cyclingflash, 29 April 2026; Cycling Weekly, 1 May 2026. The reports do not isolate the AI's contribution from training and racing.
Written with AI in the loop: my idea, AI drafted and sharpened, my judgment on the way out. Every word is mine to stand behind. — Darren