The Search Product Is Already Inside The Model

People are poking at GPT-4-turbo and finding what appears to be agentic search just sitting there — higher quality than Perplexity, not announced, not documented, discovered by accident the way you discover a hidden room by leaning on the wrong bookshelf.

Perplexity raised at a $520 million valuation in January. Google has been in search for twenty-five years. OpenAI apparently decided both of those things were fine and just shipped search as a capability inside the model anyway, quietly, without a press release.

The thing that should concern Perplexity is not that OpenAI is competing with them — it's that the competition isn't even a product. It's a feature someone noticed. There's no search team, no search brand, no dedicated infrastructure pitch. It's just the model knowing how to look things up, the same way the model knows how to write Python.

That's a different category of problem than "a better competitor launched." That's the product category itself collapsing into infrastructure.

We're going to keep seeing this. Someone builds a company on top of a capability gap. The model closes the gap. The company discovers it's been selling something that was always going to be a footnote in a release post.