He taught machines to hear at Google. Now his company, Tofu, teaches them to write.
EJ Cho runs a company named after a soft, unassuming block of bean curd, and the joke is entirely the point. Tofu takes whatever flavor you give it. So does his software: feed it your website, your decks, your half-finished campaigns, and it returns finished landing pages, emails and ads tuned to each segment you sell to. Marketers used to buy a dozen tools to do that. Cho's bet is that they will buy one.
The numbers say the bet is landing. Since launching in late 2023, Tofu has grown revenue twelvefold, signed enterprises like Check Point, Fastly and RingCentral, and pulled in roughly $17 million from SignalFire, Index Ventures and HubSpot Ventures. Not bad for three founders who, by their own admission, had never worked a day in marketing.
Start with the schools. Cho grew up a global nomad, moving through seven elementary schools on multiple continents, with stops in London and Singapore. The lesson he took from it was not geography. It was that people, wherever they sit, want roughly the same things and are annoyed by roughly the same things. And that he could handle not knowing what came next.
He turned that comfort with uncertainty into a PhD in electrical engineering at Stanford, then walked into Google's speech team in 2013. His job was to train neural networks to turn human speech into text. This was the groundwork for what the world would later call large language models, back before anyone called them that. He had a front-row seat to a revolution that had not started yet.
Then came the crisis. In his third year at Google, Cho watched Sundar Pichai walk past and felt something curdle. He calls it an early midlife crisis. The realization was simple and uncomfortable: the corporate ladder, even a very tall and prestigious one, did nothing for him. He wanted risk. As he puts it, "Not all smart people at large companies would take risks. And I wanted to take risks."
So he left to build Bolt Communications in 2018. The problem was real - speech recognition had low error rates, around 5%, but when it did slip, users had no way to fix it, just a maddening dead end. The market turned out to be too narrow to support a company. But the failure handed him a map: while selling Bolt, he kept running into how broken and fragmented marketing technology was. He filed it away.
After Bolt, he went back to big tech, joining Meta during the pandemic as it pushed conversational AI across its platform. There he met Honglei Liu, an ML leader who had done time at Twitter, Affirm and Facebook. Cho knew his own shape by then - a generalist product founder who needed someone who could, in his words, execute like "a machine." Liu was that machine.
For two years they moonlighted as market cartographers, mapping where generative AI could actually become a business. "We were constantly ideating and launching stuff. Constantly. Like, every week." The trouble was timing. The market wasn't ready for AI-first companies yet, and Cho spent the stretch wandering what he calls the AI wilderness - conviction with no validation. Then, in November 2022, ChatGPT arrived and the ground turned fertile overnight.
Validation walked in wearing a marketer's hat. Elaine Zelby, a VC-turned-operator who had built marketing teams at Slack and elsewhere, met Cho and promptly described Tofu's value back to him before he could pitch it. "She gave me my own pitch. I didn't even have to say it." She became co-founder number three.
Plenty of startups in 2023 slapped a chat box on a language model and called it a product. Cho went the other direction. "Everyone's building demos or prototypes, but very few are focused on the actual workflows or delivery." Tofu's job is not to suggest. It is to deliver - finished Google Slides, finished emails, finished landing pages.
At the center sits the Playbook, an AI knowledge graph Tofu assembles by scraping a customer's website and marketing materials, then sorting it into company facts, target segments and reusable assets. From that, it adapts copy across every channel and personalizes it per segment, plugging into HubSpot and Salesforce instead of replacing the marketer's brain.
Cho chose B2B marketing on purpose. It is drowning in text, and text is exactly what generative models do best. The pitch to buyers is blunt: "We replace and can support the multiple use cases you're purchasing individual tools for with one platform." Martech bloat, meet the off switch.
His operating philosophy is speed over polish. "You can't just wait for the perfect product - you have to keep shipping, watching how people use it, and adapting fast." And he means it about his team, too: "I'd rather you go and do something without my permission - piss off a customer, break the site - than sit there and wait for someone to tell you what to do."
Figures Tofu and its investor SignalFire cite from customer deployments. Read them as marketing claims, not audited results - but they sketch the shape of the promise.
"Not all smart people at large companies would take risks. And I wanted to take risks."
"Speed matters more than you think. You have to keep shipping, watching how people use it, and adapting fast."
"Everyone's building demos or prototypes, but very few are focused on the actual workflows or delivery."
"She gave me my own pitch. I didn't even have to say it."
"You can just start building."
"I'd rather you break the site than sit there and wait for someone to tell you what to do."