MyFitnessPal Bought Something Anyone Can Build

2026-03-03by MrPhil

MyFitnessPal just acquired Cal AI, the viral calorie-counting app built by teenagers. Cal AI hit 15 million downloads and $40 million in revenue in under two years. The deal terms weren't disclosed, but MyFitnessPal had been courting them for nearly a year, and the Cal AI team "didn't have to sell" — so it wasn't cheap.

Here's the thing that keeps nagging at me: what exactly did MyFitnessPal buy?

The Tech Is the Easy Part

Cal AI's core feature is straightforward. You take a photo of your food, an AI model estimates the calories, and the app logs it. That's it. That's the product.

Two years ago, building that required real engineering talent. Today? As I wrote about in What Is Agentic Coding, AI agents can plan, execute, test, and iterate autonomously. You could wire up a vision model API call, wrap it in a React Native shell, and have a working prototype by dinner. The AI models that power this — GPT-4o, Claude, Gemini — are available to anyone with an API key. The calorie estimation isn't proprietary. It's a commodity.

Zach Yadegari built the first version in his high school classroom. That's not a knock on him — it's a testament to how accessible the tools have become. A teenager with a laptop and an API key can now build what would have taken a funded team of engineers three years ago.

So What Did They Actually Buy?

Distribution. Fifteen million downloads. Brand recognition in the fitness space. An app that's already sitting on millions of phones with notifications turned on. A lean team that knows how to grow a consumer app.

They didn't buy technology. They bought a customer list.

And in the old playbook, that makes sense. Acquiring your fastest-growing competitor is a time-honored tradition. It's cheaper than competing, and you remove a threat while gaining users. MyFitnessPal CEO Mike Fisher said as much: "no single product can serve every consumer." Translation: we'd rather own both products than fight for the same users.

The Playbook Has a Problem

The old acquisition logic assumes that building the product is hard and acquiring users is the real moat. But AI has flipped that equation — or at least severely weakened one side of it.

If two high schoolers can build a $40M/year calorie tracking app, what stops the next two high schoolers from doing the same thing next month? The answer used to be "the code is hard to replicate." That answer doesn't hold anymore.

The real barriers now are distribution, brand trust, and data network effects. Cal AI has those. But MyFitnessPal already had distribution and brand trust. They had 20 years of it. What they lacked was the willingness to cannibalize their own product with an AI-native experience.

That's the actual story here. MyFitnessPal didn't lack the ability to build what Cal AI built. They lacked the organizational courage to ship it. So they bought it instead. It reminds me of what Stripe did right with their coding agents — the competitive advantage isn't the AI model, it's the infrastructure and willingness to actually use it.

This Hits Close to Home

I'm wrestling with exactly this question right now. I need a project that makes money, and I keep staring at my options wondering which ones will survive the next wave of AI tooling.

Take Changesmith, my SaaS that turns git history into polished release notes. It analyzes your commits, reads the actual diffs, learns your changelog's voice, and generates release notes that don't sound like a git log. It has a GitHub App, a web dashboard, a CLI, Stripe billing — the whole stack.

But here's what keeps me up at night: anyone could build their own version that covers 80% of what Changesmith does. Hook up a Claude API call, feed it a git log, and you've got passable release notes in an afternoon. The barrier to a "good enough" solution is almost zero.

Changesmith's value is in the other 20% — the style matching, the diff-aware context, the one-click GitHub Releases publishing, the webhook triggers on tag creation. The polish. The reliability across edge cases. The things that make it a product instead of a script.

But does that 20% matter enough? Is it a moat or just a speed bump?

I genuinely don't know. And the Cal AI story doesn't make me feel better. Zach's answer was to grow so fast that the question became irrelevant — he got acquired before commoditization could catch up. But that's not a strategy most indie builders can count on.

What Builders Should Take Away

If you're an indie builder or a small team, this story cuts both ways.

The good news: the tools to build competitive products have never been more accessible. A photo-to-calories app is a weekend project now. The barrier to creating something real has collapsed.

The bad news: if your product is just a thin wrapper around a commodity AI capability, your moat is paper-thin. The next kid with a laptop can replicate your core feature. Your only defense is to grow so fast that incumbents would rather buy you than compete — which is exactly what Zach pulled off.

The deeper lesson is that in the AI era, the value isn't in the build. It's in the distribution, the brand, and the speed. Zach didn't win because he had better technology. He won because he shipped fast, marketed well, and grew before anyone else in the space took AI-native seriously.

I don't have a clean answer for myself yet. Maybe the 20% is everything. Maybe it's nothing. But code is cheap now. The cost of building something and finding out is lower than it's ever been. I'd rather ship an idea and see if it gets traction than spin my wheels in analysis paralysis trying to predict which projects will survive the next wave.

MyFitnessPal's acquisition isn't a story about the power of AI. It's a story about the failure to use it — and how expensive that failure gets when someone else moves first.