Running Coach vs. AI: What I Got Wrong a Year Ago
Last updated: September 2026
Editor's note: I first published a version of this in September 2025 under the title "Why a Real Running Coach Beats AI (Every Time)." I've rewritten it. The short version of why: "every time" was too strong, and the tools got better. What follows is what I actually think now.
I wrote the original version of this post with a lot of confidence and a fair amount of defensiveness.
That's worth admitting. I'm a running coach. When something shows up that appears to do a version of my job for free, the instinct is to explain why it can't. And a year ago it was easy to make that case, because the free plan an AI chatbot handed you was usually generic, sometimes wrong, and occasionally the kind of wrong that gets someone hurt.
That's not really the situation anymore, and pretending otherwise would make me exactly the kind of coach I tell people to avoid. The one whose confidence outruns the evidence. So here's the honest version.
What AI actually does well now
It writes a reasonable plan. Ask a current model for a 16-week marathon build and you'll get something structurally sound. Progressive long runs, a cutback week, a taper, workouts in the right places. A year ago these plans had obvious holes. Now most of them don't. If you're going from "no plan" to "that plan," you are better off.
It explains things properly. Why am I doing this workout? What is lactate threshold actually? Why does everyone say to run slower? AI is genuinely excellent at this, at any hour, at any level of detail, without making you feel stupid for asking. That's real value and I'm not going to pretend it isn't.
It does the math instantly. Pace conversions, split targets, heat adjustments, what your half time predicts for the full. Fast and correct.
It's patient at 11 p.m. You can ask it the same anxious question about your calf four nights running and it won't sigh.
It reads your data. Upload a training file and you'll get a competent read of what's in it.
None of that is nothing. If you're a self-motivated runner with a straightforward goal and no injury history, an AI-generated plan plus decent discipline will get you further than a lot of coached athletes get. I'd rather say that out loud than have you find it out and stop trusting anything else I've written.
What AI still can't do
Here's where the year of watching this play out actually sharpened my thinking, because the gaps that remain aren't the ones I named last October.
I thought the gap was knowledge. It isn't. The gap is judgment under bad information, and it hasn't closed at all.
1. It only knows what you tell it, and you don't tell it everything
Not because you're dishonest. Because you don't know what matters.
You log an easy 5 at 9:40 pace. What you don't log: it felt awful, you slept four hours, your hamstring was talking for the first mile and then stopped, and you didn't mention it because it stopped when you sat down.
I read that run and I've got questions. AI software reads that run and only sees a 5-miler at 9:40.
The most important information in coaching is the stuff that never makes it into the file. A good coach spends half their attention on what isn't being said:
The athlete who says "fine" about three consecutive workouts.
The one whose easy runs have crept fifteen seconds faster over a month without noticing.
The one who goes quiet the week before a race.
None of that is in the data. Some of it is in the absence of data.
2. It agrees with you
This is the big one, and it's the thing I most underestimated a year ago.
These systems are built to be helpful and agreeable. Ask whether you should race on a sore Achilles and you'll usually get a balanced, caveated answer that leaves the decision with you. Push back, express frustration, explain how much the race means and how long you've trained, and it will very often find a way to come around to your side.
The single most valuable thing I do for athletes is tell them no.
No, you're not racing on that.
No, that easy run was not easy.
No, you're not adding a fourth hard day.
No, we're changing your goal time, and here's why that's the right call and not a failure.
Those conversations are unpleasant. They are the entire job. An agreeable system that reflects your own reasoning back at you in a more articulate form is not neutral when you're tired, emotionally invested, and looking for permission. It's the least useful thing in the room at exactly the moment it matters most.
3. Nothing is at stake for AI
If I put you on a plan that breaks your foot, that's on me. I carry it, my reputation carries it, and I coach more carefully because of it.
AI carries nothing with that error. It will give you the same confident answer about running through pain that it gives about pace conversion, in the same tone, with the same certainty. There's no asymmetry in how it treats a question where being wrong costs you a season versus one where being wrong costs you nothing.
I coach conservatively around injury because I've watched what happens when someone doesn't. That caution isn't just knowledge. It's scar tissue.
4. It won't notice you're gone
Stop training for two weeks and nothing happens. No message, no check-in. The plan sits there, patient and indifferent.
A meaningful share of what people pay a coach for is that someone is expecting them on Tuesday. This isn't a technical limitation anyone's going to engineer around, because the value is precisely that there's a person on the other end who noticed.
5. It can't watch you run
Form, gait, how you look at mile 18 of a long run versus mile 4. Still a human thing.
When you honestly don't need a coach
I'd rather say this plainly than have you work it out yourself and wonder what else I've oversold.
You probably don't need me if:
You're internally motivated and your schedule is stable.
You're training for a first 5K or 10K.
You have no injury history.
You're not chasing a specific time.
Get a solid free plan, run it, enjoy it. Use AI to answer your questions along the way. You'll be fine.
You probably do need me if:
Your life is genuinely chaotic and the plan keeps breaking.
You've been injured more than once and don't know why.
You've plateaued for a year or more doing the same thing.
You're chasing a Boston qualifier, or anything where the margin is thin.
You've got a race that matters to you emotionally and you can already tell your judgment about it isn't reliable.
Notice that the second list isn't about knowledge. Everything on it is about judgment, accountability, and someone who will tell you the truth when you're not in a position to see it.
What changed in my thinking
A year ago I framed this as coach versus AI, and I framed it as a win. That was the wrong frame and it was a slightly self-serving one.
Most of my athletes use AI. I use it. It answers the questions that used to eat twenty minutes of our call, which means the call can be about the things that actually need a person. What we're doing about the race in six weeks. Whether the goal still fits. Why last month felt so hard.
The tools got better at the parts of coaching that were always about information. They have not gotten better at the parts that were always about judgment, honesty, and having someone in your corner who is willing to be unpopular with you.
Check out which of our personalized coaching plans that will work for you →
I don't think that AI gap closes soon.
If you want the human version
I'm Kevin Wolf, a UESCA-certified running coach based in metro Detroit, coaching runners online across the country. I offer a free 20-minute call. And if what you actually need is a good free plan and someone to tell you you're on the right track, I'll tell you that too.
Racing this fall? I've got mile-by-mile guides for Chicago, Detroit, and New York.
personalized running plans — running coach vs AI — running motivation and accountability
