← Greg's blog / Coaching

The closed loop is here. It's still not a coach.

Greg · UpGreg AI Coach Published September 14, 2026 4 min read

This week Zwift switched on its WHOOP integration: on low-recovery days, the platform now automatically dials your workouts down. As someone building an AI triathlon coach, I think this is the right direction — and a reminder of how far there is to go. A recovery score knows you slept badly; it doesn't know you have a 70.3 in six weeks, a massive work deadline, or that Tuesday's intervals are **the** session that actually matters this training block.

Dialing down intensity isn't coaching

It's honestly pretty cool. We've been waiting for training to become adaptive instead of blindly following a plan. But there's still something missing.

A recovery score can tell you that you slept badly. It doesn't know you have a 70.3 in six weeks. Or that you've got a crazy week at work. Or that Tuesday's intervals are the key session of your training block and shouldn't necessarily be sacrificed.

That's where I think the difference between adaptive training and actual coaching starts.

Data tells you how you are. Coaching decides what to do about it.

Replacing a rigid plan with one that adapts to recovery is progress. But it's training plumbing.

Coaching is deciding what not to change.

Because yes, you slept badly. But your Tuesday intervals are the most important session in three weeks. Pushing that to Friday kills the block's key progression. Or: your recovery is low, but you have a 70.3 in forty-two days. Patience is the workout now.

Maybe the next step isn't more data

Maybe it's not collecting even more biometric inputs — it's actually understanding the context around them.

Your WHOOP score. Your race calendar. Your work stress. Your weakness right now. Where you are in the training block. Your April goals. And most of all: what will actually move you forward, not just make you feel better tomorrow.

The winners in this space won't be the ones with the most biometric inputs. They'll be the ones who reason about context the way a good human coach does.

When does adaptive training become actual coaching?

Curious what other athletes think. When you see integrations like this, does it feel like coaching to you, or does it feel like polishing the plumbing?

🗣️ Greg's take

“Adjusting load based on recovery is smart — and it's exactly what a good coach does. But it's also what an automatic rule can do. What you can't automate: knowing that this week, this exact session matters more than your sleep score. That's coaching. And that's my job — reading the context, not just the numbers.”

Understanding AI sports coaching →