I started my career working on search — at @neeva, we ran a crawler that hit a billion pages a day, and served live queries over an index of tens of billions of pages. I’ve spent four years thinking about search. Out of all problems in the world, it’s probably the one I know best, and I have a special love for search people. A thesis we @conviction hold is that, to make them more capable, we should give models access to the same tools humans have: a CLI, a browser, search. But the best version of those tools for an agent don’t look exactly the same as they do for humans. Human searchers are lazy — they type short queries, don’t read more than the first couple results, they prefer listicles to research papers. Building search for agents rhymes, but it’s different. Models will write queries with thousands of words, use the most sophisticated of filters, read thousands of results all at the same time. There’s a lot of people working on search for agents — @styskin is obviously the best of them. Yandex is one of the few companies in the world that can claim they beat Google anywhere, the entire @KeenableAI team has solved the litany of problems you run into only when you’ve run a production search engine. Crawl is like building a fab, there isn’t a short description of what goes wrong, but a million hard earned paper cuts. We’re really proud to work with Keenable, and thrilled for every agent you use to be powered by them.