The Most Important AI Products Won't Come From Labs

Everyone’s watching the frontier labs. The reasoning benchmarks, the context windows, the multimodal capabilities. And yes, those matter.

But I think we’re misreading where value actually accrues.

The labs are competing on capability. They have to—that’s the game they’re playing, and they’re playing it well. But capability stopped being the bottleneck somewhere in the last eighteen months. Google’s Gemini 3 (Flash or Pro model, take your pick) was “good enough” for most things that matter to most people. Everything since has been incremental from a user’s felt experience.

The actual bottleneck now is intimacy. Products people want to return to. Products that feel like they know you. Products that occupy a slot in your life rather than a tab in your browser.

Big companies are structurally incapable of building these. Intimacy requires being opinionated, weird, and willing to let some people hate you. It requires a point of view that can’t survive a committee. The returns to personality are about to be enormous, and personality doesn’t scale the way infrastructure does.

This is the window. The models are commoditizing. The wrappers get mocked, but the wrappers are where the actual relationship lives. The labs will keep publishing benchmarks while the value capture happens at the edges—small teams building things that make people feel something.

I’m not saying capability doesn’t matter. I’m saying it’s table stakes now.

The question isn’t “who has the best model?” anymore.

It’s “who builds something people actually love?”