Search should be free, especially for AI agents

Exa has a good team. They are a search and indexing company that started in 2021, before the AI wave, then pivoted into agentic search and got distributed early into the LLM builders that mattered.

Credit where it's due, but none of that changes the fact that search shouldn't be metered by the token: it was free for people, and it should stay free for the agents doing the searching now.

Tell a two-year-old she can only ask why three times a day. She doesn't run out of questions, she learns that questions cost something, and stops asking why the sky is blue, why she can't eat dirt, why anything at all.

Or, imagine Google charging for the results, a fraction of a cent per blue link. You'd feel it immediately: less clicking, less searching, less wondering. That's exactly the world token-metered search is quietly building for agents, one API call at a time.

Here's the case on why we should imagine a different world, and how.

Rationed search is worse search

If search costs money per call, you do one of two things: you search a smaller corpus, or you take whatever ranking the provider hands you, sponsored results included, and either way you've quietly agreed to search less than you need.

The persona changed. The pricing didn't.

Google made search free by monetizing the user instead of the query, which was fine. But humans only ever read the top ten results on page one, because that's what a human has patience for, and billions of pages sat there unread, not because search was constrained, but because we were.

Now the searcher is an agent, with unlimited time and unlimited patience, no fatigue at result eleven and no impatience at page twelve, and we finally built something with the capacity to search the way search was always supposed to work.

Here's why the industry won't let that grow. Agents don't fall for dark-pattern ads, they don't impulse-buy what they didn't need, and you can't monetize an agent's attention the way you monetized a human's. So instead of ads, the industry is taxing curiosity, charging per query and hoping somebody pays.

Both sides of search, the searcher and the corpus, could finally run unconstrained at the same time, for the first time ever. Instead, we freed one and metered the other. Imagine what search looks like when neither side is rationed. That's the world we're arguing for.

What you lose when you ration curiosity

Give a two-year-old three questions a day and tell her that's her limit, and she stops asking why.

Kids ask why relentlessly because nobody's charging them for it, and adults answer relentlessly because that curiosity is the whole point. Search worked the same way, until we decided agents should pay by the query. Cap the breadth, rerank to the top ten to save cost, and you've rebuilt the exact limitation humans had, this time on purpose, in software.

Why this is TinyFish's fight

TinyFish exists so that everything on the web, big or small, shallow or deep, open or behind a login, gets found when it's its turn to be found. A free search world is our mission.

Of course, we aren't a charity

If all you sell is search, you have to charge for search. That's the box Exa is in, along with every agentic search startup launching almost weekly.

In the old world, humans did the work of both the searcher and the harness carrying them around the web. In the new world, agents do that work. They don't want a list of links, they want the outcome. Once the corpus is built, that's where the value accumulates, not in the lookup that got you there.

The stack we get paid for sits above that, not inside it: fetch, login and authentication, at scale cloud browsers, agent navigation good enough to trust, and Mako, the model we built specifically to operate the web instead of describing it. Contextual graphs and memory come next, so an agent that visited a site last week walks back in already knowing it.

Proxy rotation and CAPTCHA solving live inside that stack because a browser has to survive the modern web to run at all, the same way it has to speak TCP. Nobody buys TCP: they buy the browser, and everything it lets them do next.

Infrastructure has always worked this way: DNS, NAT, routers, all once visible and now invisible. Order headphones on Amazon tonight, and a Cisco router earns its fee somewhere along the path. You'll never see the invoice or think about it.

Our search will get paid, when the agents deliver the entire outcome, accurately.

Giving search away pays for itself. It's the structure that lets search be what it was always supposed to be: free, wide open, safe to ask why as many times as you want.

This is the right thing to do, and at this point, it's also the only thing to do.