The Search Engine That Thinks the Internet Should Be Built for Machines, Not Humans

SAN FRANCISCO — May 22, 2026 — William Bryk has a prediction that sounds absurd until you realize the data already backs it up. Sometime this year, he says, AI agents will conduct more searches on the internet than humans. Not someday. Not in a decade. This year. “There are billions of humans doing searches,” he told Bloomberg this week. “There are going to be trillions of agents, like, very soon.”

Bryk is the CEO of Exa Labs, a five-year-old San Francisco startup that just closed $250 million in funding at a $2.2 billion valuation—more than tripling its $700 million price tag from last fall. The round was led by Andreessen Horowitz, the firm co-founded by Marc Andreessen, whose own Netscape browser helped humans navigate the early web in the 1990s. The symmetry is deliberate: Andreessen bet on the browser that let humans explore the internet. Now his firm is betting on the search engine that lets machines do it instead.

Exa is part of a wave of startups racing to redefine online search for an era in which the dominant user is no longer a human typing queries into a box, but an AI agent scouring the web on a human's behalf. The incumbents—Google, with its $2 trillion market cap built on a quarter-century of search dominance—have watched the ground shift beneath them. OpenAI's ChatGPT demonstrated that conversational AI could answer questions directly, threatening to make Google's list of blue links obsolete. Now the next wave is arriving: AI agents that do not merely answer questions, but act on them—booking flights, comparing insurance policies, researching legal precedents—and they need a search infrastructure that was never designed for them.

Exa is building that infrastructure. Its search engine is not designed for humans. It is designed for AI agents—structured, high-quality, machine-readable data that agents can ingest, process, and act upon without ever showing a human a list of links. The company has landed thousands of customers, including AI coding startup Cursor, enterprise AI platform Cognition, and CRM giant HubSpot. The number of queries flowing through its systems has grown from roughly 100 million per month in April 2025 to approximately 1 billion per month in April 2026—a tenfold increase in a single year.

The Agent-First Architecture

To understand why Exa exists, one must understand what an AI agent needs from the web—and how profoundly that differs from what a human needs.

When a human searches for "best restaurants in Chicago," Google returns a page of links—Yelp reviews, Eater articles, OpenTable listings—and the human clicks, reads, compares, and decides. The search engine's job is to retrieve relevant documents. The human's job is to do the rest. This architecture has defined web search for twenty-five years. It works brilliantly for humans. It is nearly useless for AI agents.

An AI agent tasked with booking a dinner reservation in Chicago does not want a list of links. It wants structured data: restaurant names, addresses, hours, reservation availability, price ranges, cuisine types, and recent reviews—all in a machine-readable format that it can process instantly. It wants this data without pagination, without advertising, and without the JavaScript-heavy interactive elements that make modern web pages difficult for machines to parse. It wants the web, stripped of everything designed for human eyes.

Exa's search engine is built from the ground up for this use case. It crawls, indexes, and structures web data specifically for consumption by AI agents. Its APIs return clean, structured results that agents can ingest and act upon. The company operates its own $5 million GPU cluster for processing and indexing, while also running significant workloads on Amazon Web Services. The architecture is capital-intensive—the $250 million round will more than double its workforce and expand its computing capacity—but the thesis is that the market for agent-first search will dwarf the market for human-first search.

The partnership with Google announced in April 2026—under which Google's Gemini model can access Exa's search engine, which is specifically designed for AI agents—is either a validation of Exa's approach or a hedge by Google against the obsolescence of its own search architecture. Bryk declined to disclose whether the deal includes a financial component, but the strategic signal is unmistakable: the dominant search company of the human-web era is already wiring itself into the agent-web era.

The Competitive Landscape

Exa is not alone in the race to build agent-first search infrastructure. The broader ecosystem of AI-native search startups has attracted significant venture capital, reflecting a market that is reorganizing around the assumption that traditional search is ripe for transformation.

Tavily, another AI search startup, has built a platform optimized for retrieval-augmented generation. TinyFish has focused on real-time web indexing for agentic workflows. Parallel Web Systems, led by former Twitter CEO Parag Agrawal, recently raised $100 million at a $2 billion valuation from Sequoia Capital, pursuing a similar thesis: that the web must be reindexed and restructured for consumption by machines.

The incumbent, Google, is not standing still. The company has integrated its Gemini model directly into Search, launched AI Overviews that synthesize answers from multiple sources, and built agentic capabilities into its ecosystem. But Google's structural challenge is the same one that faces every incumbent during a platform shift: its existing business model—advertising against human search queries—generates hundreds of billions of dollars annually and cannot be abandoned overnight. The transition to agent-first search, in which humans never see a results page and therefore never see an ad, is existentially threatening to Google's core revenue engine. The company must manage the transition carefully, preserving its cash cow while building its replacement.

Exa has no such conflict. It does not sell advertising. It sells API access to structured web data, charging developers and enterprises for the volume of queries their agents consume. The business model is aligned with the technology transition rather than threatened by it. The company's growth from 100 million queries per month to 1 billion per month in a single year suggests that the market for agent-first search is expanding faster than even optimistic analysts predicted.

The Andreessen Connection

The involvement of Andreessen Horowitz in Exa's funding round carries symbolic weight that extends beyond the dollar amount. Marc Andreessen, the firm's co-founder, helped create the first widely adopted web browser—Netscape Navigator—in the 1990s. That browser was the interface through which millions of humans first experienced the internet. It was built for people: graphical, clickable, designed around human cognition and human curiosity.

Now, Andreessen's venture firm is making a fresh bet on how people—and increasingly, agents—will interact with the web. The investment is a recognition that the browser era, the search-engine era, and the human-directed-web era are giving way to something new: a web that is navigated primarily by machines, on behalf of humans, using infrastructure that was purpose-built for the task.

Bryk has been careful to frame Exa not as a Google competitor, but as a new category of infrastructure. "It's humans using agents using search," he said. The formulation is precise: humans are still the ultimate beneficiaries, but they are no longer the direct users. The agent is the intermediary, and the agent needs a different kind of search engine than the one Google built.

What This Signals

The Exa story is not primarily about a startup raising $250 million at a $2.2 billion valuation. It is about the structural transformation of the internet's most fundamental function—search—from a human-facing service to a machine-facing utility.

For twenty-five years, Google's search engine has been the primary interface between humans and the world's information. It has been so successful, and so dominant, that "to Google" became a verb in dozens of languages. But that dominance was built on an assumption that is now being eroded: that the primary consumer of search results is a human being, sitting at a screen, reading links and making decisions.

The rise of AI agents—autonomous systems that can book meetings, research legal precedents, compare insurance policies, and execute complex multi-step tasks without human intervention—is changing that assumption. An agent does not need a list of links. It needs structured, machine-readable data, delivered instantly, at a scale that dwarfs human search volume. Exa's tenfold query growth in a single year—from 100 million to 1 billion per month—is an early indicator of how fast the shift is happening.

The Andreessen Horowitz investment is a bet that the shift is structural, not cyclical. The Google partnership is an acknowledgment by the incumbent that the agent-first architecture cannot be ignored. The $2.2 billion valuation is the market's preliminary judgment on whether a startup built for the agent-web era can carve out a durable position alongside—or perhaps eventually, instead of—the search giant that defined the human-web era. The trillions of agent queries that Bryk predicts are coming. The question is which search engine they will use.