I've studied dozens of early-stage startups building AI-native products, and most of them won't survive the next five years. Not unless they're doing three specific things. So before the next model update decides it for you, let me show you exactly what those are.
The day $285 billion vanished
I watched it happen in real time. When Anthropic shipped Claude Cowork, roughly $285 billion in software market cap evaporated in a single trading session. Thomson Reuters had its worst day on record. Within a week, the cumulative damage neared $1 trillion.
What's crazier is how one product update did that, to companies that were sure they were safe. But the market didn't punish everyone the same. It spared the ones that owned their data and their customers. That told me exactly which companies die on the next release, and which ones don't.
The three layers a model can't copy
So here's what the survivors had in common.
Own your data
They owned data that only existed because of their product. Every workflow and every interaction their users made compounded into something nobody could copy with a bigger model or more money. The ones that didn't capture that data had no moat. They just hadn't noticed it yet.
Own your distribution
They owned a way to reach people. A better model doesn't reach customers on its own, and the harsh truth in this market is that something better always ships. The companies without a distribution channel, without an audience that trusted them, watched their users leave for the cheaper alternative.
Own your vertical
They owned a vertical nobody could shortcut. General models are exactly what the name says: general. They don't know the specific regulations or edge cases of your industry. The founders who survived knew that because they lived in it, while the ones building thin and wide got punished.
The test: if the best model got 10x better tonight
So here's the test I run on every one of these companies now.
If the best model on earth got 10x better tonight, what do you still have tomorrow?
If the honest answer is "nothing they couldn't rebuild in a weekend," you're just waiting to be absorbed.
The fix: build depth, not speed
The fix isn't to build faster than the new models come out. It's to own the three layers they can't touch: data, distribution, and expertise. Building with real depth in a vertical is the only thing that's actually worth anything right now.
So make sure you clear all three. And if you missed one, that's exactly what to fix next week.
