The Honest Read
Here's where the AI training pitch breaks down. Coaching is not a plan-writing exercise. The plan is the easy part. The hard part is watching a power file and noticing that the rider's threshold work has gone flat at the same time the chat tone has shifted, and pulling the next block back before they tip into overreaching. AI tools don't have that signal. They have what you typed.
Generative AI is genuinely useful for things it's good at — drafting an explanation, summarising a study, suggesting a session variant when you're stuck. None of that is coaching. Coaching is the conversation around the plan. It's the email at 9pm on a Tuesday saying "skip Thursday's intervals, the data says you're cooked, here's why." It's the 20-minute call after a bad race that ends with a plan tweak you'd never have spotted yourself.
The other thing AI tools don't do well: nuance under fatigue. A good coach knows that the same workout on paper looks completely different in week 4 of a build vs week 1 of a recovery week. They know when a rider says "that felt fine" they actually mean it was a struggle. They know which athletes need pushing and which need pulling back. AI plan generators give every athlete the same template structure with a few variables swapped.
There's one place AI is fair game — as a learning tool. If you're self-coached and you want to understand why a coach has prescribed something, ask an AI to explain the physiology. If you want a session variant when your usual climb is closed, AI will give you ten options. Don't let it run your training. Use it to learn — then either apply that knowledge as a self-coached athlete or hand the keys to a human who's been doing this for years.
FAQ
Why doesn't AI work for cycling coaching?
AI plan generators only know what you tell them. They don't pick up tone, hesitation, recovery patterns over months, or the small signals in a power file that an experienced coach has seen a thousand times. Cycling coaching is judgment built on years of working with athletes — that's not a thing you replicate with a prompt.
Can AI replace a coach for budget reasons?
If the budget genuinely doesn't allow coaching, an AI plan beats no plan — but a structured app like TrainerRoad or a community programme like Roadman's Not Done Yet will usually beat a generic AI plan for the same money. AI works best as a learning aid alongside structured training, not as the structure itself.
What about apps with built-in AI coaching?
Adaptive apps (TrainerRoad, JOIN Cycling, others) are different from raw AI tools — they have curated workout libraries, validated plan structures, and measurable adjustment logic. They're a meaningful step up from a ChatGPT prompt. They're still not a coach. They adjust workouts inside a plan; they don't rebuild the plan based on the human you are.
Is there any role for AI in serious cycling training?
Yes — as an explainer, a session generator, and a way to interrogate concepts. Use it to learn the physiology, draft an analysis question, or build a session variant when you're stuck. Don't use it to run your season. Coaches across the sport, including most coaches we've had on the podcast, are using AI as a tool inside their workflow — not as a replacement for the work.
How do I know I've outgrown AI plans?
Three signals: gains have stopped despite consistent effort, your plan keeps getting derailed by life and you can't rebuild it well, or your event is non-standard and the AI keeps producing generic blocks that don't match it. Any of those, you've outgrown algorithmic training. Time for a human.