Microsoft Made a Bing for AI Agents

Microsoft predicts that, soon, most online searches will be completed by artificial intelligence agents acting on behalf of humans, rather than humans themselves. And the big tech company has just announced Microsoft Web IQ, its new suite of APIs built for AI agents that helps them scrape the web faster and more comprehensively.

Web IQ is a search engine for AI systems, Microsoft said during its annual developers conference, Microsoft Build, on Tuesday. It connects AI tools to information across the web, but presents it in a different way than a search engine would for a human.

“We traditionally have had search engines for humans, like Bing,” Jordi Ribas, Microsoft’s president of search and AI, told CNET during an interview on Monday morning. 

Ribas said what’s rapidly growing now is search engines for agents. “Web IQ is our solution for this type of search engine, and it provides this context … the web documents, news, images, videos that are relevant for an agent query, and then they’re used by the agent for grounding.”

While humans and robots search differently, Microsoft was able to lean on its expertise from Bing to build Web IQ.

“We’ve leveraged everything that we’ve done in the last 20 years from Bing, while at the same time re-architecting and rebuilding it from the ground up to make sure we have a very optimized search engine for agents,” Ribas said.

The difference between search activity by humans versus AI agents comes down to how the information is presented. For humans, that means ranking search results according to what the person is trying to find. For AI agents, it looks like a comprehensive but condensed page of results to parse without using too many tokens. So Web IQ had to deliver high-quality results quickly, in a compact package.

Tokens are small bits of written text — usually around four characters in the English language per token — that AI models process to form responses to your prompts. The more tokens the AI model uses, the more expensive the output is.

Agentic AI is the next step beyond chatbots. Instead of an AI telling you which steps to take, it steps in and completes tasks on your behalf. You may have already dealt with some forms of agentic AI, like self-driving cars and smart home assistants, but the latest AI agents, like OpenClaw, can operate far more independently.

Ribas said Web IQ leads the industry across search result quality, token efficiency and latency — 95% of the time, it responds in less than 165 milliseconds — and in token efficiency. Microsoft said it’s around 2.5 times faster than any other product out there.

The Web IQ APIs, which are a component of Microsoft IQ — also announced during the Microsoft Build keynote — have already been used for powering both Microsoft’s own AI offering, Copilot, as well as OpenAI’s ChatGPT “for quite some time.” 

“Many other systems are also grounded with Web IQ,” he added, though Microsoft won’t currently name which.

Ribas said that while he thinks the number of searches done by agents will be lower than some estimates — which suggest they’ll be making 1,000 times more queries than humans within a few years — he thinks AI agents will certainly be making more than people by the end of the year.

“We anticipate that agents will easily do more queries than humans, just by the nature of how many queries they do for every chat response, and the fact that obviously AI systems are going to continue to accelerate and grow,” Ribas said. “And for every query or every problem that those systems run, generally you need multiple queries done in a system like Web IQ.”