Of late, I have had some conversations with people whose bank balances I would quietly trade for.

Older folks, unfailingly humble, and generous with how they got there. None of them were in tech. Their work sounds like car showrooms, spice trading, and chemical additives for industries I cannot pretend to understand. Businesses that have grown quietly for twenty, thirty years, while the rest of us were busy reading about whatever came out yesterday.

I have been thinking about those conversations against the steady drumbeat of my social media feed, where some version of "the AI window is closing", "you have to move now" shows up twice a day.

It is true. It is also a bit misleading.

There isn't one AI window. There are at least three, and most of the panic is tuned to the wrong one.

Window 1 is the foundation model layer, run by Anthropic and the likes. If you weren't in this race two years ago, you aren't getting in now.

Window 2 is the AI-native application layer. Cursor, Perplexity, Harvey, etc. A very crowded and currently hot category, where most of the movement and most of the noise is.

Window 3 is AI applied to the industries currently too unglamorous for the Valley to notice. Regional logistics. Mid-market manufacturing. Vertical software for seemingly mundane categories. This window is barely cracked open, and could stay that way for a long time.

The numbers on this are striking. Over 90% of small and mid-sized industrial businesses have no AI in place at all and only about a quarter of family businesses in India treat digital and new tech as a real priority.

The people I had coffee with are not racing anyone. They have been building slowly so far, in categories that slip under the radar, and the wealth has quietly followed. They did all this before AI existed, and the next generation of them, I suspect, will do the same with AI as a quiet operational lever rather than a headline. A spice trader running better demand forecasts on a Thursday afternoon. A dealership where the customer calls finally stop waking someone's cousin at midnight.

None of it will look like the AI you scroll past, and that is rather the point.

The story of the next decade is going to be less about models and more about operators, using these models to re-invent industries no one really talks about. Quite a few of these operators, going by my coffees, are doing very nicely already.

The people telling you to hurry are mostly building in Windows 1 and 2, where speed is the whole game. If your edge is operational depth in a category Silicon Valley finds boring, their urgency is not your urgency.

The Three AI Windows