Frontier labs and the retreat from consumer software

April 15, 2026

Frontier LLM companies are starting to remind me of old computer companies.

Early on, the software experience itself is the trademark. The product is memorable because the interface feels alive, surprising, and newly possible. But over time, the center of gravity shifts. What matters more is the platform, the enterprise layer, the API surface, the control layer, the operating-system position.

Once that shift happens, high-care consumer surfaces start to get deprioritized. Not necessarily because they are technically impossible, but because they are no longer the most important layer in the company’s strategic stack. Voice intimacy, niche UX, persona-heavy products, even things like NSFW modes, begin to look less like core bets and more like edge surfaces.

This is why some of the most obviously desirable AI consumer experiences still feel oddly unfinished. The capability may exist, but the organizational attention has moved elsewhere.

Part of why this pattern feels familiar is that it resembles an older computer-company dynamic. A company can first become culturally important because the software feels magical, personal, and sharply differentiated. Later, as scale grows, it concentrates more on the substrate, the operating-system layer, and the distribution advantage. The narrower, stranger, or more obsessive application layer gets left to third parties.

That is part of why I keep thinking about Apple here. Not because the analogy is exact, but because the structural motion feels similar. The company that defines the stack is not always the company that cares most about every beloved surface on top of that stack.

That creates an opening. The most loved AI consumer products may increasingly come not from the foundation model companies themselves, but from startups that obsess over one experience and iterate on it with real care. The labs may own the substrate. But the products people actually love could be built one layer above.

In that sense, the future may look familiar. Just as the most important computer companies were not always the ones that made the most beloved software experiences, the most important AI labs may not be the ones that define AI consumer software either.