AI Coach for Cyclists: Power-Based Training, FTP, and What the Data Can Do

Cycling hands AI coaching the best raw material in endurance sport. A runner's pace bends with hills, wind, heat, and GPS error. A power meter measures the work itself, at the crank or the pedal, in watts, the same way every ride, which means 250 watts on a cold Tuesday in February is the same 250 watts on a July climb. Same number, same work. That repeatability is what a coach, human or AI, needs before any plan means anything.

So the question for a cyclist is never whether the data is good enough. It is. What gets done with it is the open question, because your head unit and your platform of choice already compute most of the classic numbers. An AI coach earns its place in the layer above the math: reading those numbers against your strength work, your sleep, your food, and your actual week, then adjusting the plan and telling you why.

Below: power as the signal, what the device ecosystem already handles, what it structurally cannot, what an adaptive cycling plan looks like in practice, and the honest limits.

The short version. AI training for cyclists is worth it when your week is unpredictable, when you lift as well as ride, or when you want the plan to answer to your recovery data instead of a calendar. It is not worth it if you only want zone math. Your head unit already does that part, for free, and does it well.

Power is the signal

One anchor holds up everything in structured cycling: FTP, the highest power you can hold for roughly an hour. Your zones come from it. So do your interval targets and your load scores. Test it with a 20-minute protocol or estimate it from recent rides, then retest every 6-8 weeks in a build, and the FTP calculator will run the standard estimate while the power zones tool turns the result into watt ranges for each zone.

Two derived numbers do most of the analytical work after that.

Normalized power answers what a raw average cannot, namely how hard the ride cost you: a punchy group ride with surges and coasting might average 180 watts but cost your body what a steady 230 would. Same file, two different truths. The weighting math sits in the normalized power explainer, and you can run the calculation on any ride yourself.

Training load stacks those ride costs into a picture over weeks: one number per ride, scaled by how hard it was relative to your FTP and how long it went. Sum it. Track the rolling total and fatigue shows up in the numbers before your legs report it, and the Training Stress Score guide covers the mechanics and what the rolling numbers mean.

A third model sits underneath both, for riders who race short and hard. Critical power describes the boundary between the effort you can sustain and the effort that burns a finite reserve above it, and it comes from two or three all-out tests at different durations, which lets it predict what you can hold for any time in between. Pacing a 10-minute climb, a pursuit effort, anywhere FTP alone is too blunt. The critical power tool runs the model from your test results.

None of this requires an AI. It requires arithmetic and a power meter, which is exactly the point: cycling hands a coach clean, quantified inputs that most sports cannot.

What your head unit already does, and where it stops

Ride with a Garmin and the device ecosystem computes a lot on its own. Auto-detected FTP. Load ratios. VO2 max trend. Recovery time after each ride. Daily suggested workouts that shift with your overnight HRV. For a cyclist who rides, looks at the suggestion and follows it, that free layer does real work, and the Garmin AI coach guide goes deep on what it does well.

The ceiling is structural. Three parts.

It cannot see training it never records. Squats on Monday never enter the ride-load model. They surface as unexplained stress in your HRV, which the algorithm reads as noise, so three heavy gym sessions plus a long ride registers as one hard day. For a cyclist doing serious strength work alongside the bike, the watch is planning from half a picture.

It cannot ask you anything. The algorithm produces a session type, and it has no idea your race moved up three weeks, that your left knee complained on the last climb, or that work travel eats Thursday. You interpret the suggestion against your life. It never interprets yours.

It has no cross-domain view. Ride data, sleep, and HRV live in the model. Your food does not. Neither does your gym log. Your last conversation about goals does not exist, because there was no conversation.

Deterministic power math is a solved problem. Context is not. That line is where an AI coach starts.

What an adaptive cycling plan looks like

An adaptive plan is one that gets rewritten when reality diverges from the template. Four situations come up for every cyclist. That is where the difference shows.

FTP moves. Six weeks into a build your threshold has crept up, static plans keep prescribing the old watts, and your "threshold" intervals quietly drift down into sweet spot while the stimulus softens. An adaptive coach catches the signal in your recent rides, prompts a retest or adjusts the estimate, and rewrites the zone targets so hard days stay hard. Race where the road tilts and watts per kilo matters more than raw watts. The power-to-weight tool puts a number on it.

You miss sessions. Work runs long, a kid gets sick, it rains for four days. A template leaves a hole and moves on, while a coach redistributes: the key interval session survives, an endurance ride shrinks, and the week's load target gets recalculated instead of abandoned. The skill in coaching missed weeks is knowing which session was the point of the week. Volume is replaceable. The quality stimulus usually is not.

Your body pushes back. HRV trending below baseline for three mornings, resting heart rate up, sleep short. The consensus on morning readiness signals is clear enough to act on: push threshold work into a suppressed state and you buy fatigue, not fitness. So an adaptive coach downgrades Tuesday's intervals to zone 2, holds the pattern for a day or two, then reschedules the quality work for when the trend recovers. Read the HRV-guided training guide for the framework over weeks.

Indoor and outdoor are different rides. An hour on the trainer has no coasting, no descents, no traffic lights, so the same interval set costs more inside, and a plan that treats trainer hours and road hours as interchangeable overloads the indoor weeks. Small adjustment. It compounds across a winter.

The numbers stay the same numbers through all four. FTP, normalized power, load. What changes is that something reads them daily, against everything else in your life, and rewrites the week instead of waiting for you to notice.

The honest limits

An AI coach does not improve on the power math. Load scores, normalized power, FTP detection from ride files: deterministic arithmetic, and your device ecosystem already computes it precisely. An AI reasons about those numbers. It does not calculate them better. Anyone claiming otherwise is selling something.

Bike fit is human territory. Saddle height, cleat position, reach: hands, a trained eye, and you on the bike in front of someone. No data stream substitutes.

Pack skills too. Cornering in a group, reading a break, holding a wheel at 45 km/h: you learn that on the road, in groups, over years, and a coach can structure the fitness underneath it. The craft is yours to build.

Chasing a national-level result, or managing a complicated injury? Hire a human coach. The relational accountability of a person who notices you went quiet, plus the judgment earned from hundreds of athletes, is worth the money at that level. For the cyclist training 5-12 hours a week around a job, the AI coach is the practical call: available every morning, reads more data than a human has time for, and costs less per month than one coaching consult.

What Movement Rebels does for cyclists

Concrete, so you can judge it against the list above.

Garmin Connect sync, both directions. Rides, power data, HRV, sleep, and training load flow in, and out the other way the coach pushes structured workouts back to your Garmin: warm-up, intervals with watt targets, cooldown. You press start. The device runs the session. On iOS, the native app also reads Apple Health.

A debrief after every ride. Each completed synced ride triggers an automatic coach analysis in your chat thread, a lap and power breakdown in plain language: where the power faded, how the effort compared to the plan, what it means for tomorrow. One paragraph. Not a dashboard.

Weekly plans from the full picture. Writing your week, the coach reads your last 14 activities, your recovery data, your strength log, and a persistent athlete profile built from every conversation you have had with it. Fueling reaches it the way you write it: put "cutting on 300-500 kcal" in Notes for your coach and it reads a climbing resting heart rate against that context. It does not see your meal log. Tell it your knee is sore once. It remembers when it programs your next block.

The rest of the stack in the same app. A strength library of 3,900+ workouts for the gym days. Rebel Fuel sets macro targets that follow the day's session, so carbs scale up when the ride is hard and settle back on rest days. Power makes the fueling side of cycling unusually precise: work in kilojoules converts to calories burned at close to one-to-one, which is why a power file beats any heart-rate estimate for calculating ride calories. Recovery tools round it out. And a morning brief reads your overnight data, then tells you what today should look like before you decide anything.

A social layer, at a smaller scale. Crews, a feed, follows and high fives are in the app. Segments are not, and neither is Strava's number of riders. Strava sync exists for that gap: your rides land in your friends' feeds automatically. Keep Strava for the reach. Run the training here.

You can ask questions. "My FTP test felt terrible, was it the heat?" "Race moved to five weeks out, what do we cut?" "Should the long ride move to Sunday?" The coach answers from your data, in your thread, with the context of everything above.

Triathlons on your calendar? The triathlon coaching guide covers the multi-sport version. For the broader comparison against hiring a person, read AI coach vs personal trainer.

FAQ

Can AI write a cycling training plan that actually works?

Yes, when it is working from power data. Cycling is the easiest endurance sport to program for, because a power meter measures the work itself rather than estimating it. The plan quality depends on what the system does with FTP, load and recovery signals, not on the AI label. A tool that only generates sessions from a template is doing something different from one that rewrites next week after reading your last ten days.

Do I need a power meter for AI training to be useful?

No, but it changes what you get. With power, the coach works from a repeatable measure of effort and can set interval targets in watts. Without it, everything falls back to heart rate, which lags the effort, drifts with heat and fatigue, and makes short intervals nearly impossible to prescribe precisely. A smart trainer indoors gives you power for the sessions that need it most.

How is an AI coach different from the training suggestions on my Garmin?

Your Garmin plans from what it records: rides, sleep, HRV. It cannot see your gym sessions, it cannot ask why you skipped Thursday, and it has no memory of your goals. An AI coach reads the same ride data plus your strength log, your fueling and the conversation history, then explains what it changed. The zone math is the same either way. The context around it is not.

How often should FTP be retested?

Every six to eight weeks during a build block is the common recommendation. Test more often and you spend training days on testing. Test less often and your interval targets drift out of date, so threshold work quietly becomes sweet spot. Some platforms also estimate FTP from recent hard rides between formal tests.

Is AI coaching enough on its own, or do I still need a human?

For a cyclist training five to twelve hours a week around a job, an AI coach covers the programming, the adjustment and the analysis. Hire a human when you are chasing a national-level result, managing a complicated injury, or when you know you need a person who notices you went quiet. Bike fit and pack skills stay human work regardless.

Pricing

Movement Rebels is free for 7 days: full coach, full Garmin sync, full strength library, Rebel Fuel, recovery tools, no credit card. After the trial, $20/month.