AI Perspectives
Beyond the foundation model: building moats in the application layer

If everyone can call the same model, where does defensibility come from? Data, workflow depth and distribution in unconventional niches.
When every startup can access the same frontier models through an API, the model itself is no longer a moat. Defensibility has to come from somewhere else.
We see three durable sources. First, proprietary data generated by the product's own use, which competitors cannot buy. Second, deep integration into a customer's workflow, so that switching means rebuilding processes. Third, distribution into a niche that larger players ignore because it looks too small or too strange.
The strongest companies combine all three. They start in an unconventional market, earn the right to collect unique data, and use that data to make their agents measurably better than any general-purpose tool.
Frequently asked questions
What makes an AI startup defensible?
Proprietary data from product usage, deep integration into customer workflows and strong distribution in a specific niche make AI startups defensible.