Founders building in AI keep treating it as a technology competition. It isn't. The underlying models are commoditising faster than anyone predicted, and the moat is somewhere else entirely.
Where the moat actually is
What matters now is who owns the distribution, the data, and the workflow. The companies that will win aren't the ones with the best model — they're the ones embedded deeply enough in a customer's daily operations that switching becomes genuinely painful.
The Uber parallel
I saw this pattern at Uber. The technology was never the durable advantage. The network was. The trust was. The habit was. By the time competitors had comparable technology, Uber had years of behavioural data and market presence that couldn't be replicated quickly.
The question worth asking
For AI companies building today: the question isn't "how good is your model?" It's "how deep is your integration?" Workflow depth creates switching costs. Switching costs create defensibility. The model is rented; the depth is yours.
The founders worth backing in AI right now are the ones who can answer this clearly: why will your customer still be using this in five years, even if a better model exists? If the answer is the model itself, that's not enough.