Google's search overhaul, announced at its developer conference this week, is the most visible part of a broader shift. The company redesigned its search box for the first time in 25 years, built AI tools that browse the web on users' behalf around the clock, and pushed AI-generated summaries to the top of results pages. The effect on website traffic is already measurable. More than half of all US Google searches now end without anyone clicking through to a website. Some sites have seen traffic drop 20 to 40 percent since AI summaries became standard.
This creates an obvious problem for businesses that depend on Google to send them customers. It also creates an obvious opportunity for anyone who can build the replacement infrastructure.
Exa Labs is one of the clearest bets in that direction. The company does not compete with Google for human users. It builds a search engine designed specifically for AI agents: software programs that browse, research, and gather information automatically. When a chatbot answers your question about suppliers, or an AI assistant compiles a market report, something has to retrieve that information from the web. Exa wants to be that layer. The company raised $250 million this week at a $2.2 billion valuation, led by Andreessen Horowitz. That is more than three times its valuation from just eight months ago.
Parallel Web Systems, founded by former Twitter CEO Parag Agrawal, works in the same space. It offers web search and research tools specifically built for AI agents, with named customers including legal AI platform Harvey, workplace tool Notion, and financial data firm Opendoor. Parallel raised $100 million led by Sequoia at a $2 billion valuation, five months after raising its previous $100 million at a $740 million valuation. The speed of that jump tells you something about how urgently investors are trying to get into this category.
The practical difference between these tools and a normal search engine is worth understanding. Google was built to show results that humans find satisfying, which means it rewards content designed to attract clicks. Exa and Parallel are built for AI programs that need precise, factual information fast, with no interest in clickbait or search-optimized filler. One investor noted that Exa delivers, for a few dollars, what companies had previously spent hundreds of thousands to assemble manually.
For non-technical businesses, the implications run in two directions. First, if your company depends on Google search traffic to bring in customers or leads, the ground is shifting faster than most annual plans account for. Gartner projects that 25 percent of organic search traffic will shift to AI chatbots and assistants by the end of this year alone. Second, these search tools are becoming part of how AI software inside businesses works. An AI system that helps your procurement team research suppliers, or helps your sales team find prospects, is almost certainly pulling web data through a service like Exa or Parallel. Knowing that this layer exists, and that it is being built right now, matters when you are evaluating any AI vendor or building any AI-assisted workflow.
The bigger picture is a race to own the infrastructure that AI relies on. Google is protecting its advertising business while trying to build AI search at the same time, which creates a real conflict. OpenAI controls the most-used AI interface but cannot make search its primary focus. That leaves room, and apparently enormous investor appetite, for dedicated players to own the search layer that AI agents depend on. Whether Exa, Parallel, or someone else wins that position is genuinely open. But the category is real, the money is serious, and the timeline is short.