He walked into a Y Combinator interview in 2021 and opened with: "We are better than Google." He meant it. Four years later, Benchmark handed him $85M to prove it.
There's a specific kind of obsession that drives certain founders. Not "I spotted a market gap" obsession. The other kind - the kind where a problem follows you home, shows up in your dreams, and won't leave until you've either solved it or burned out trying. Will Bryk has that kind.
By 2021, Bryk was a software engineer at Cresta, building real-time AI products, doing solid work. He was also, in whatever hours the evenings gave him, writing a book about the history of civilization. Not a startup blog post. A book. The project forced a conclusion he couldn't unsee: the future of civilization depends almost entirely on the quality of information people consume. And the quality of information people consume depends almost entirely on search.
He looked at Google - the engine that shaped what the internet had become, the system that trained two billion people to phrase their curiosity as three-word keyword fragments - and thought: this is already obsolete. The world was building large language models that could understand the nuance of full sentences, multi-part questions, implicit requirements. Google was still matching keywords against an index built for a pre-AI world.
So Bryk left Cresta. He called his Harvard roommate Jeff Wang, who had been building data infrastructure at Plaid. They'd been friends since freshman week. In Summer 2021, they walked into Y Combinator as Metaphor Systems, and Bryk opened with: "We are better than Google." They got in.
Google has actually limited people's imaginations when it comes to what you can find on the Internet.
- Will Bryk, CEO of ExaThe technical bet was specific. Traditional search engines match keywords. Metaphor - later renamed Exa - would train a transformer model not to predict the next word in a sentence, but to predict the next link that a human would choose to share. It's a subtle distinction with enormous consequences. When someone writes "the best explainer on transformer architectures," Exa doesn't look for pages with those words. It predicts what a thoughtful person with deep context would choose to bookmark and pass along. Bryk calls it "neural PageRank on steroids."
The company found its market faster than anyone expected. AI startups building on large language models kept running into the same wall: their models knew things up to a training cutoff date and invented things after it. They needed real-time, high-quality, verified information from the web. Google's API was built for humans typing queries into a box, not for software agents asking 10,000 targeted questions per minute. Exa was built for exactly that.
Cursor - the AI code editor - became a customer. So did private equity firms and management consulting shops, running deep research on companies using the Exa API. Bryk watched the use cases multiply and saw something: the number of searches generated by AI systems was about to exceed the number generated by humans. This wasn't a niche B2B tool. This was infrastructure for the next era of intelligence itself.
By September 2025, Benchmark led an $85M Series B at a $700M valuation - ten times the Series A price from just one year prior. Peter Fenton joined the board. Nvidia's venture arm, NVentures, was in. Lightspeed stayed in. The company had 120 employees, a $5M cluster of 144 H200 GPUs for training, and thousands of developers building on its API. The company had also quietly built Websets - a product that turns web searches into structured grid-format databases, letting analysts run market research the way software runs data pipelines.
What makes Bryk unusual in a field full of intense founders is the specific way he talks about ambition. He doesn't frame the work as hustle. He calls it play. "Building something from scratch with friends," he has said, "is the most beautiful experience you could have in life." It's a claim that might read as Silicon Valley posturing, except that he backs it with operational choices. The Exa office has nap pods - sourced from China. Bryk naps every other day and credits the practice with producing his best ideas. He doesn't preach grind. He builds systems that make thinking better.
He is also comfortable with slowness where most founders fear it. His view on search latency is almost heretical: "People think searches take 500 milliseconds because we've been conditioned. But what if a search takes a minute, 10 minutes, or a whole day? That's fine." The insight is about quality, not speed. For an AI agent doing research on a company before a high-stakes meeting, a search that takes 10 minutes and returns exactly the right 10 documents is worth more than one that takes 300 milliseconds and returns 10 million indexed results tuned for ad clicks.
When hiring, Bryk looks for people who "don't care about the status quo, who can think from first principles about everything." He is not hunting for domain experts with established playbooks. He wants people who will look at the way things have always been done and ask whether that's actually right. It is, not coincidentally, the same question he asked about Google.
"We are better than Google."
"The number of searches from AIs are going to exceed the number of searches from humans."
"AGI is going to need search."
"Google has actually limited people's imaginations when it comes to what you can find on the Internet."
"Building something from scratch with friends is the most beautiful experience you could have in life."
"What if Google was outdated? What if you could use transformers to build perfect search over all the world's information? In 2021, this sounded crazy."
Engineering intern at SpaceX while still at Harvard. Interns reportedly couldn't approach Elon Musk directly.
Earned A.B. in Joint Computer Science and Physics. Co-President of Harvard Robotics Club, overseeing 100+ members and multi-year robot-building projects.
Joins the ML startup building real-time AI for customer conversations. Spends evenings writing a book on the history of civilization. The book leads to a conclusion that changes everything.
Leaves Cresta with Harvard roommate Jeff Wang. Opens YC pitch with "We are better than Google." Raises $5M seed. Builds a transformer model that predicts links rather than keywords.
Exa 2.0 launches as a state-of-the-art neural search engine. The company begins attracting major AI platforms as customers, including Cursor.
Rebrand to Exa. Launches Exa 2.1 claiming best-in-class neural search. Launches Websets - structured database generation from web searches.
Benchmark leads. Nvidia doubles down. Lightspeed stays in. Peter Fenton joins the board. 10x valuation increase in 12 months. The bet is paying off.
While engineering AI products at Cresta by day, Bryk spent his evenings writing a book about the history of civilization. The project forced a conclusion he couldn't escape: civilization's future depends almost entirely on the quality of information people can access. That conviction drove him to quit and start Exa.
Bryk sources nap pods from China for the Exa office and takes one himself every other day. His reasoning isn't about wellness as a perk. He credits the practice directly with creative problem-solving: "Closing my eyes helps me come up with creative solutions." The pods became a symbol of how Exa thinks about work.
During the earliest days of Exa, the core team lived together. It wasn't a hippie commune - it was a conscious choice to accelerate alignment and trust. The company's culture of unusual closeness and directness traces back to that house-share period.
As Co-President of Harvard's Robotics Club, Bryk oversaw eight-year-long projects where student teams built robots that could walk, roll, swim, navigate mazes, fly autonomously, play soccer, and rove Mars-like terrain. The scale of ambition involved in building physical machines from scratch - over years, with volunteer students - seems to have calibrated his threshold for what "hard" actually means.
When Google's search query for "shirts without stripes" returned poor results, Bryk used it as proof of concept during the early Metaphor pitch: a language model could understand that query intuitively. Google could not. That T-shirt query became a demonstration of the gap he was building to close.
Bryk and Jeff Wang - Exa's co-founder - met as Harvard freshmen and have been best friends since. That decade-long friendship is the foundation of how they run the company: with humor, directness, and a willingness to challenge each other that most CEO-cofounder pairs never build.
Interned at SpaceX in 2016. Interns were reportedly not allowed to approach Elon directly.
Was writing a book about the history of civilization while working as an AI engineer. The book convinced him to start Exa.
Takes naps every other day and has installed nap pods in the Exa office. Credits them with his best ideas.
Built a mini search engine in college at Harvard - years before founding the company that would become Exa.
Exa's GPU cluster - 144 H200s - cost approximately $5M to build and operates for training and search R&D.
Exa's "Twitter Wrapped" - a year-in-review product for X users powered by Exa search - went viral in 2024.