🚴‍♂️🏃‍♀️ Algorithm meets real biology: Why wearables and AI won't replace the coach – but make him more important

7-minute read | Deep dive for cycling, trail running & ambitious age-groupers

Hi! 👋

You wake up, look at your phone – and before you really know how you feel, your watch apparently already knows.

Recovery 42. HRV below baseline. Sleep score moderate. Training readiness low.

The plan was actually to do 5 x 5 minutes of VO₂max today. However, yesterday was a long workday, the child woke up twice, and my legs don't feel too bad... at all.

So you ask the AI:

"Should I train today or rather recover?"

And this is precisely where one of the most exciting topics in modern endurance sports begins.

Because technically, wearables can now do an impressive amount. Oura, WHOOP, Garmin & Co. measure heart rate and HRV with astonishing accuracy in some cases. AI systems can create training plans, analyze exertion data, and rewrite workouts in seconds. 🔬🤖

Therefore, the crucial question is no longer:

"Can technology plan training?"

She's been able to do that for ages.

The better question is:

"Can she understand what this data means for this particular person today?"

And that's where it gets interesting.

1) The good news first: Wearables are not just toys 📈

Despite all the criticism of recovery scores, one thing should not be overlooked: Sensory perception is improving.

A validation study from 2025 compared Garmin Fenix 6, Oura Gen 3 and 4, Polar Grit X Pro, and WHOOP 4.0 with reference measurements taken during the night. Oura and WHOOP, in particular, showed a high correlation with the reference data for HRV; Oura Gen 4, for example, achieved a correlation of r = 0.99.¹

That's impressive.

But – and this is the crucial point – a well-measured value does not necessarily mean a good training decision.

HRV is a signal from the autonomic nervous system. Sleep duration is information. Resting heart rate is information. Skin temperature is information.

In contrast, a readiness score is already an interpretation of this information – and the exact weighting of the individual parameters is often proprietary in commercial systems.

The 2026 guidelines on HRV research state this quite clearly: Wearable-based HRV can be useful, but it is not a direct measure of recovery, sympathetic activity, or performance readiness. The authors caution against automatically equating easily measurable variables with the truly relevant ones.²

👉 Coach's perspective: Good data reduces uncertainty. It doesn't eliminate it.

2) “Readiness” sounds precise – but is surprisingly vague from a scientific perspective 🧠

The word " readiness" is everywhere now.

Garmin Training Readiness. WHOOP Recovery. Our readiness.

The suggestion behind it: Today you are either ready – or you are not.

However, a recent systematic review from 2026 paints a different picture: In 14 studies examined, "readiness to train" was defined very differently. HRV, resting heart rate, jump tests, submaximal tests, and subjective reports on sleep, stress, fatigue, and muscle soreness were used. However, the ability of these methods to predict actual daily performance was limited or often not tested at all.³

This is extremely important.

Because your body doesn't have a central "readiness switch".

You can:

• mentally tired, but muscularly fresh,

• have a low HRV and still a good interval session

• They slept well, but are coming down with an infection.

• Being tired after a tough block of training – and this fatigue can be a desired outcome in training methods.

A score compresses complexity.

That's convenient. But that's precisely where information gets lost.

3) Subjective feeling is not the enemy of data

A common assumption is that athletes are too biased regarding RPE, fatigue, or daily form to be able to meaningfully assess themselves.

It's not quite that simple.

Of course, bias exists. Ambition, expectations, fear of missing training, or the desire to absolutely stick to a plan can influence our self-assessment.

But subjective data is therefore not worthless – on the contrary.

A study published in 2025 followed 20 international endurance athletes for a year, combining wearables, sleep, glucose, and training data with mood-state questionnaires and subjective readiness assessments. The result: Objective and subjective data complemented each other, but the relationships were highly individual . The authors explicitly advocate for a multidimensional approach rather than a single score.⁴

Or, to put it more simply:

Your feelings are a measure. They just need context.

Kilian Jornet aptly describes his own approach: He often guides training based on RPE and intuition, while simultaneously collecting a lot of data to understand why something feels the way it does. If you only look at data, you "lose the big picture"—but going solely by intuition also means missing out on information.⁵

That's a surprisingly modern training philosophy.

Not data or feeling .

Rather:

Data + feeling + experience.

4) And what can AI really do now? 🤖

We should remain fair here as well.

AI is no longer at the level of "here are three generic running units".

A pilot study from the University of Innsbruck in 2025 showed that AI-generated training plans can now be so plausible that experienced coaches could not reliably distinguish whether a plan had been created by an AI or a human. 25% of participants in structured training programs were already using AI-based plans.⁶

That's no small result.

An analysis of marathon training plans from various major language models in 2026 also showed that the systems often accurately reflect key principles such as volume progression, tapering, and a predominantly low-intensity distribution. At the same time, significant weaknesses were identified regarding advanced athletes, specific pace targets, and the differentiation between performance levels.⁷

And a study with elite swimmers and coaches found a similar pattern: AI was useful for simple, low-intensity sessions, but more frequently produced inappropriate repetition lengths, volumes, or rest periods in more complex intensity zones. Only around 28% of the athletes considered the plans usable with minor adjustments.⁸

The pattern is quite clear:

AI excels at:

• structure

• Pattern recognition

• Data aggregation

• Variants

• rapid plan changes

• standardizable problems

AI becomes weaker in:

• conflicting signals

• complex periodization

• incomplete data

• Injury and illness logic

• psychological patterns

• long-term learning about a person

• Decisions under uncertainty

• Several sports such as flat running, trail running, road and off-road cycling, and perhaps also strength training.

And that's exactly where coaching begins.

5) The most dangerous mistake: confusing correlation with cause ⚠️

Imagine:

Your HRV drops for three consecutive days.

The app detects: bad trend.

The AI detects: bad trend.

But why?

Perhaps because of:

• Training load,

• Alcohol,

• late dinner,

• Sleep deficit,

• psychological stress,

• incipient infection,

• Heat,

• Trip,

• cumulative fatigue,

• or simply normal biological fluctuation.

The fact that A occurs together with B does not necessarily mean that A is the cause of B.

And now the training methodology level comes into play.

Perhaps the current fatigue is not a mistake at all.

Perhaps we are at the end of a planned training block. The body is meant to be stressed. This will be followed by three light days and a period of supercompensation.

If a system automatically responds to every red score with "Easy Day", it can destroy the very stress logic it is supposed to optimize.

Of course, modern systems are more complex than a simple traffic light. But the fundamental question remains:

Does the system only know your state – or does it also understand the intention behind that state?

6) This is precisely where the coach's true value lies 🎯

A modern coach should not be in competition with technology.

If a coach ignores wearables, performance data, AI or TrainingPeaks in 2026, that is, in my view, just as outdated as an athlete who blindly leaves everything to an algorithm.

The coach sits one level above .

It connects four levels:

1. Data: What are we measuring?

2. Context: What happens in life and during the training phase?

3. Interpretation: What part of this is signal – and what part is noise?

4. Decision: What do we do today and why?

Abbildung: Wichtigkeit von Coaches bei der Nutzung von Wearables und KI, eigene Darstellung (KI generiert).

That sounds trivial. But it's the very core of professional training management.

Tadej Pogačar describes his collaboration with his coach Javier Sola less as a secret algorithm than as daily communication, trust, and more specific preparation for upcoming races.⁹

Jonas Vingegaard even said of his long-time coach Tim Heemskerk that he was "more than just my coach" and difficult to replace. Heemskerk himself describes coaching as an applied science and art , implemented with experience and intuition.10

Interestingly, even UAE Team Emirates is now working intensively with integrated data and AI. Performance Director Jeroen Swart describes the combination of performance, sleep, temperature, wellness, medical, and blood data as a major advancement. AI generates recommendations from this data – but the final training control is still carried out by a performance team.11

This is probably the model of the future:

AI does not replace coaching. AI changes good coaching.

7) The difference is rarely noticeable on a perfect Tuesday.

If your life is stable, you are healthy, and the plan is working, a good app can be fantastic.

Tuesday: Intervals. Thursday: Sweet Spot. Saturday: Long Ride.

No problem.

The value of a coach becomes more apparent on these days:

Thursday, 6:30 a.m.

Sleep score 82. HRV okay. Legs heavy. Conflict at work. The most important training session of the block is on Saturday.

What shall we do today?

Or:

Two weeks before the A-race

Slight scratchy throat. Excellent form. TSB/Readiness perfect. Athlete is eager to train.

How much risk are we willing to accept?

Or when trail running:

Three weeks before the Ultra

Fitness level is good. Uphill performance is strong. But after downhill sessions, the quadriceps muscles remain tired for four days.

The clock shows training load.

The coach sees:

The race will be decided on the downhill section. We have a specific resilience problem.

This is precisely where data analysis becomes training control.

8) Does every ambitious athlete need a coach?

No.

And that's important to me.

If you:

• you train regularly

• you are healthy

• have fun

• you achieve your goals,

• listen carefully to your body,

• and your daily life is relatively stable,

A smart platform or a well-prompted AI system can be perfectly sufficient. 👍

A coach becomes particularly interesting when:

• your time budget is limited

• you have a clear performance goal,

• Training often clashes with work or family life,

• You don't know what to change despite having a lot of data,

• Injuries or overuse injuries recur,

• you constantly “run over” yourself,

• or you need to be in top form at exactly the right time.

Then you're not primarily buying a training plan.

You buy better decisions .

9) My aecoaching approach: Hybrid instead of holy war 🤝

I don't see AI, wearables, and modern analytics platforms as competition.

I use them.

Performance data, HRV, subjective perception of exertion, TrainingPeaks, performance diagnostics, trends and AI-supported analysis can be extremely valuable.

But no single value gets the last word.

Because behind every data set is a human being.

With a job.

Family.

Motivation.

Uncertainty.

Ambition.

And sometimes a sleep score of 48 – even though he feels surprisingly good. 😉

Therefore, for me, the future of coaching is not man versus algorithm.

She is:

Algorithm + sports science + experience + relationship.

Conclusion: Miss a lot. But understand more. 🔬🧠

Wearables are getting better.

AI is getting better.

And that's a good thing.

Because the better our tools become, the less time we need to spend collecting data – and the more time we can spend on the really difficult question:

What does this mean for you now?

Kilian Jornet sums it up perfectly:

" Data helps us understand what lies behind our feelings. But if we completely ignore feelings, we lose the big picture ."⁵

That's exactly what aecoaching is all about.

No more data for the sake of data.

But better decisions for your athletic development.

🚀 Interested in coaching that uses your data – but sees you as a person?

If you're an ambitious cyclist, runner, or trail runner, and you feel that while Garmin, TrainingPeaks, Strava & Co. are measuring more and more about you, they aren't automatically making your training decisions any easier, then we can take a look at it together.

At aecoaching.at we combine sports science training management, data analysis and psychological coaching into a system that is designed to fit your sport and your life .

👉 A no-obligation initial consultation is not a sales pitch, but a brief assessment: Where are you now, where do you want to go – and would coaching even be beneficial for you?

More information at www.aecoaching.at

Sources & further reading 📚

Dial et al. (2025). Validation of nocturnal resting heart rate and heart rate variability in consumer wearables. Physiological reports.

Guidelines for rigor and reproducibility of heart rate variability within human cardiovascular research (2026). American Journal of Physiology – Heart and Circulatory Physiology.

Conceptual Frameworks and Assessment of Readiness to Train in Endurance Exercise: A Systematic Review (2026).

Validating Subjective Ratings with Wearable Data for a Nuanced Understanding of Load-Recovery Status in Elite Endurance Athletes (2025).

5. Kilian Jornet (2026), Contribution to RPE, Data and Interpretation; supplementary COROS Training Data Feature (2025).

6. Wachholz F. et al. (2025). Acceptance and trust in AI-generated exercise plans among recreational athletes and quality evaluation by experienced coaches. BMC Research Notes.

7. Artificial intelligence-generated marathon training programs: reliable tools in exercise prescription for athletic performance? (2026).

8. The role of artificial intelligence in sports training: opportunities, challenges and future applications for competitive swimming (2026).

9. Pogačar T. (2025), Interview about Coach Javier Sola, Cycling Weekly.

10. Heemskerk T. / Vingegaard J. (2026), Interview about long-term coach-athlete collaboration, Cycling Weekly.

11. Swart J. (2026), Interview about integrated data and AI use at UAE Team Emirates-XRG, CyclingNews.