A hobby project,
shared because it might help.

How it started

I love golf. I’m not great at it. I played a round in February 2026 and played awful — the kind of round where you walk off wondering if you should quit the game entirely.

I’d watched what feels like a hundred YouTube videos. Read the books. Tried the tips. Walked off the range every Saturday the same golfer I’d walked on as. The information was never the problem — there’s an endless amount of it. The problem is nobody is watching me practice and telling me the one thing to fix.

After that round, I started thinking about whether AI could help. I’d been using Claude for work and figured I could try to vibe-code something — feed it a swing video, see if it could tell me what to actually work on.

It worked better than I thought. The first version was rough — me feeding swing videos to Claude and asking it to read the pose data. But even that early cut found things my eye had been missing. After a few weekends of evening hours, I had something I was using on the range. It remembered what we’d talked about between sessions. It picked one thing to work on instead of fifteen.

I showed it to some friends. They wanted it too. So I cleaned it up, gave it a name, put it online.

What this is, and what it isn’t

This is a hobby project. I have a full-time job, a young family, and a finite number of evening hours. I work on Coach Harvey between bath time and bedtime, on flights, in the gaps. Some weeks I ship a lot. Some weeks I don’t ship anything. That’s the deal.

I’m basically the target customer. I’m not going to drop $200 on a lesson every week. My local pro is overbooked anyway. I built this for me first — and then for everyone in the same spot, which I’ve come to realize is a lot of people.

It’s free for now and will stay that way as long as I can manage. If that ever has to change, I’ll be honest about why. I’m not trying to compete with your coach. If you have one you trust, use them. Coach Harvey is for the rest of us.

The hard part

The hard part wasn’t the AI. That’s the easy part now — Claude can read pose data, find faults, hold a conversation. The hard part was figuring out what a coach actually does, and getting the model to do that one thing well.

Most swing apps tag fifteen faults at once. I get why — the model finds fifteen things wrong, so it lists fifteen things. But that’s not coaching. A coach picks one thing. Maybe two. Then watches you work on it until it sticks. Most of the hard work on this has been getting Coach Harvey to stop dumping observations and start picking the right one. A lot of iteration. A lot of throwing my own swings at it and arguing with the output.

The technical version — how Claude reads pose data, the 44-fault library, how full swings vs chips vs putts route through different algorithms — lives at How it works.

What’s working, what isn’t

Fault detection on full swings is solid — driver, irons, wedges, the seven-phase model holds up well. The chat that reads from your swing history is the part I’m most proud of. You can ask “what should I work on this week” and get an answer that’s grounded in what Coach Harvey has actually seen from you. That feels like the real unlock.

Rough spots: the putt and chip pipelines are newer than the full-swing one and don’t catch as much. Some of the tools (yardbook, bag audit, practice plans, debrief) ship in waves — they all work, but some are crisper than others. The video upload can take a beat. And occasionally Claude calls a fault that isn’t really there — most of my evenings go into chasing those down. I’m working on all of it.

Named for Harvey Penick, who taught more golfers than anyone in the game. Harvey Penick’s Little Red Bookis the bestselling sports book of all time — a coach pulling players aside, one piece of advice at a time. That’s what I was trying to build.

Feedback, suggestions, bugs, “this doesn’t work for me,” “I tried it and stopped because of X” — all welcome. Especially the last one. Email me at ian@coachharvey.ai. I read everything.