An AI training coach reads your training history, your overnight recovery data and your calendar, decides whether the session you planned still fits the day you are having, and changes the workout itself when it does not. The change lands in the structured workout file, so it reaches your watch, head unit or indoor trainer app like any other session.
That is the short answer. The rest of this guide explains how it works in practice, using IntervalCoach's daily process as the worked example: what the coach reads, the checks it runs each morning, and the limits that shape how the method is used.
What an AI coach reads
Three kinds of information go into every decision.
Training history and load. Every completed session adds to a running picture of fitness, fatigue and form, the balance between the two. This is what a plan is built from and what tells the coach how much work you have absorbed recently.
Overnight recovery. A wearable measures what your body did while you slept: HRV, resting heart rate and sleep, and on devices that record them, skin temperature, breathing rate and blood oxygen. Each signal is compared with your own recent baseline, because a resting heart rate that is normal for one athlete is elevated for another.
Your own input. Races, group rides and sessions you planned yourself sit on your calendar. A short morning check-in records how you feel, and after each session you can rate how hard it was. These carry as much weight as the sensors: a number that says you are recovered does not override you saying you are not.
Five checks, every morning
Before a session reaches you, IntervalCoach runs five checks.
1. Health. If you have marked yourself sick or injured, the day becomes rest with recovery guidance. Coming back, training resumes gradually: hard sessions return step by step, the reduced load lasts longer after a longer illness, and it lifts early once your recovery signals are back to your own normal.
2. Calendar. An A or B race on the day means no extra session is added, and you get race-day guidance instead. A group ride already on the calendar counts as that day's training. Sessions you planned yourself are always kept.
3. Readiness. The coach weighs your overnight signals, your recovery from recent sessions and your check-in into a picture of how ready you are. Recovery is modelled individually: IntervalCoach fits a recovery curve to each athlete from their own mornings, so it learns how quickly you in particular come back from a given kind of session. When you are still inside the recovery window of a hard session, the next day is held below threshold, so a planned VO2max or threshold day becomes tempo or endurance.
4. Reshape. If the day does not match the plan, the workout changes. The load can come down, the intensity can be capped at a zone, the stimulus can be swapped, the warm-up can be extended, or a breathing block can be added to a longer cooldown. Several changes can apply together, and the most restrictive combination wins.
5. Generate and steer. The finished session is written as a structured workout to your calendar, where it syncs to your devices. Across the week the coach keeps you on your training load target: if yesterday ran well over, today eases; if you are behind, the remaining days rise within safe limits.
The full process, including what happens when overnight data arrives late, is covered in how IntervalCoach adapts your workout every day.
The plan the day adapts from
Daily adjustments only make sense against a plan. IntervalCoach builds weekly plans around one of five periodization models: linear, undulating, polarized, pyramidal and block. The model sets the shape of each week, how many hard sessions it holds and how intensity is distributed, and the daily checks then fit each session to the day. The differences between the five are explained in five periodization models explained.
What it gets you
The practical result is a session that fits the day you are having. A poor night before a planned threshold session can change that session before you start it, so the adjustment happens at the point where it can still help.
It also keeps a plan intact when life interrupts it. An illness, a group ride added to the calendar or a race that moves is absorbed into the week, and the load target for the week steers the days that remain.
And it shows its reasoning. Every change is recorded with the reason it was made in the Agent Log, so you can see what the coach saw and disagree with it when you know something it does not. If you want something different, you can ask for it in your own words in the Coach+ chat.
Limits that shape how it is used
These limits are part of the design, and each one decides how a signal is used.
Readiness informs the day. It never overrules you. Overnight data is a useful signal and a noisy one on any single morning. That is why a recovery estimate is one input among several and never overrides what you report about how you feel.
Early signs of illness are acted on, never diagnosed. In IntervalCoach's analysis of athlete-logged sick days, resting heart rate climbing over several days was the earliest reliable sign. The coach eases training in response, and a rise in resting heart rate never triggers that on its own, because a hard training block can produce the same pattern.
It can only read what your device records. Skin temperature and breathing rate sharpen the picture, and not every wearable measures them. The checks run on whatever signals are available.
Recovery is shared across your sports and judged per sport. Swimming, cycling and running draw on the same recovery, so a hard session in one sport affects the next day in all of them, while each sport keeps its own zones and its own tolerance for accumulated fatigue.
It works alongside a coach as well as without one. A human coach can see the same dashboard as the athlete, edit the weekly plan and join the chat, and their instructions carry into the sessions the system writes.
Questions worth asking any AI coach
If you are comparing AI coaching tools, four questions separate them quickly.
- Does it change the workout, or only send advice? An adjustment that lands in the workout file reaches your device. Advice in an app is easy to miss before a session.
- Can it tell you why? A coach that states the reason for each change lets you judge it.
- Does it compare you with yourself? Baselines built from your own data read your signals correctly. Population averages misread the athletes who differ from them most.
- Does it respect what you put on the calendar? Your races, group rides and planned sessions should shape the week around them.
Frequently asked questions
What is AI coaching for endurance athletes?
AI coaching is software that plans and adjusts training from an athlete's own data. It reads training history, overnight recovery measurements such as HRV, resting heart rate and sleep, and the athlete's calendar, then decides whether the planned session still fits the day. When it does not, the coach changes the structured workout itself, for example by reducing the load, capping the intensity or swapping the session, and syncs the result to the athlete's watch, head unit or indoor trainer app.
What data does an AI training coach use?
Three kinds. Training history and load, including fitness, fatigue and form. Overnight recovery data from a wearable, such as HRV, resting heart rate and sleep, and where the device records them, skin temperature, breathing rate and blood oxygen. And the athlete's own input: races and planned sessions on the calendar, a morning check-in, and how hard each completed session felt. In IntervalCoach each overnight signal is compared with the athlete's own recent baseline.
Can I see why an AI coach changed my workout?
In IntervalCoach, yes. Every adjustment is recorded with the reason it was made, so an athlete can see what the coach saw before the session and disagree with it when they know something the data does not show. Changes can also be requested in plain language through the Coach+ chat.
How does an AI coach change a workout?
It edits the structured workout file before the session reaches the athlete. In IntervalCoach the adjustments include reducing the training load, capping the intensity at a given zone, swapping the session for a different stimulus, extending the warm-up, adding a breathing block to an extended cooldown, or replacing the session with a rest day. Several adjustments can apply at once, and the most restrictive combination wins. Because the change is in the file, it syncs to the watch, head unit or indoor trainer app like any other workout.
Can AI coaching detect illness?
It can see early signs in overnight data, but it does not diagnose. In IntervalCoach's published analysis of athlete-logged sick days, a multi-day rise in resting heart rate was the earliest reliable sign, with HRV dropping mostly in the last couple of days. The coach uses that pattern to ease training toward recovery, and a rise in resting heart rate never triggers that response on its own, because a hard training block can produce the same signal.
Does AI coaching replace a human coach?
It covers the daily adjustments a coach would make from data, every morning and before each session, which suits self-coached athletes. It also works alongside a human coach: in IntervalCoach a coach can see the same dashboard as the athlete, edit the weekly plan and join the chat, and their instructions carry into the sessions the system writes.