Shouldn't Coaches Be Coachable Too?

The short version: Skepticism about AI in coaching is healthy when it means asking why a suggestion was made and whether it matches what the athlete actually said. It turns harmful when "I've done this for 20 years" becomes a reason not to look at information that could help an athlete. The standard we hold athletes to — listen, try things, change when something isn't working — should apply to coaches too.

---

I failed.

At least, that's how it felt when my college swimming career ended. I'd worked for years to swim Division I at Colgate. Then I tore my pec, and I left the sport. Most of what I cared about had been tied to swimming, and for a while I didn't know what came next.

Eventually I started coaching. Except now someone else was counting on me. People were trusting me with their health and their goals, and I didn't want to let them down.

Here's something I think is hard to admit: you can care a lot about your athletes and still miss something.

The athlete who said it twice

Imagine you have 30 athletes. One of them says a workout felt harder than usual. Later that week, they say it again. Their times look fine. You have a whole practice to run.

A month later, you read back through their feedback and realize they've been struggling for longer than you thought.

Nobody did anything wrong in that story. The athlete told you. You heard them. But two comments, a few days apart, buried in a week of 30 people's comments, don't look like a pattern when you're standing on deck. They only look like a pattern when you put them next to each other.

I'd want something that helped me notice that sooner. Even if all it did was put those comments side by side so I could ask a better question.

Where my skepticism kicks in

But if a tool suggested changing that athlete's training, I'd need to know more.

• Why that change?
• Did it consider what the athlete actually said, or just the numbers?
• Could I explain the decision to the athlete without pointing at a screen?

That last one matters most to me. If I can't explain a training decision in my own words, I don't really own it. And if I don't own it, I shouldn't be making it.

That's the kind of skepticism that protects athletes. It isn't "AI bad." It's "show me your reasoning," which is the same thing a good coach asks of a good assistant coach.

Two ways to get it wrong

I don't think being a good coach automatically makes you good at using AI. There are two failure modes, and they point in opposite directions.

You trust it too much. You accept a bad suggestion because it saves time. Researchers have a name for this — automation bias — and it shows up everywhere from aviation to medicine. People with a confident-looking recommendation in front of them tend to stop checking it, especially when they're busy (Parasuraman & Manzey, 2010).

You trust it too little. You refuse to look at useful information because you're convinced you already know your athletes. This one has a name too: algorithm aversion. People who see a tool make one mistake often stop using it entirely, even when it's more accurate than they are on average (Dietvorst, Simmons & Massey, 2015).

Most conversations about AI in coaching only talk about the first one. I think the second one is just as real, and coaches are more likely to fall into it, because our identity is built on knowing our athletes.

"I've been doing this for 20 years"

Those years count for a lot. Experience is how you know that a kid who goes quiet before a meet is nervous, not tired. No dashboard is going to replace that.

But "I trust my experience" needs questioning too. You still have 30 athletes, one practice to run, and only so much you can keep in your head. Experience tells you what to look for. It doesn't give you more hours to look.

As an athlete, I measured failure through my own performance. As a coach, I worry about missing something that matters to someone else's. I wouldn't want my confidence in what I know to get in the way of learning something that could help them.

What being coachable looks like for a coach

I've been debating accountability in my AI Ethics course at NYU, and one question keeps coming up: who is responsible when AI causes harm? The coaching version of that question has a less-discussed flip side — who's responsible when you don't use a tool that would have caught something?

I don't have a clean answer. But I think a coachable coach does a few things:

1. Uses AI to notice, not to decide. Let it surface patterns — repeated comments, a slow drift in sleep, a missed check-in streak. Keep the decision.
2. Demands the reason. Any suggestion that can't explain itself in terms of what the athlete said and did shouldn't change their training.
3. Checks the tool against reality. If it flags someone, go talk to them. Over time you'll learn where it's useful and where it isn't.
4. Notices their own defensiveness. If your first reaction to new information is "I already know," that's worth a second look.

The question I keep coming back to

When does skepticism protect your athletes, and when might it keep you from helping them?

We ask athletes to listen, try things, and change when something isn't working. If new evidence challenges how we've always coached, shouldn't we be coachable too?

---

This is part of why we built Matter the way we did: it reads training, readiness, and athlete check-ins, puts the relevant context in front of the coach, and drafts changes with their reasons. Nothing is written until the coach approves it. You can see how that works in the worked example.

References

• Parasuraman R, Manzey DH. Complacency and bias in human use of automation: an attentional integration. Human Factors. 2010;52(3):381–410.
• Dietvorst BJ, Simmons JP, Massey C. Algorithm aversion: people erroneously avoid algorithms after seeing them err. Journal of Experimental Psychology: General. 2015;144(1):114–126.