When Eunjoon “EJ” Cho tried to market his first startup, he acquired a second job. There was the product to build, certainly. Then there were the tools required to tell people about it. Each channel seemed to come with its own software, vocabulary and small examination in whether he knew what he was doing. The browser tabs were multiplying faster than the answers.
The experience stayed with him. Cho had spent years around artificial intelligence and engineering, including work at Google, Meta and Affirm. Yet here was an ordinary business task that remained irritatingly difficult. A company could buy plenty of software and still leave a person responsible for stitching the work together. That person, for a while, was him.
In 2023, he co-founded Tofu with Honglei Liu and Elaine Zelby. The company makes AI software for B2B marketing teams, connecting the creation of personalized content across channels. Its origin becomes easier to understand if you start with the founder facing his own marketing chores. He had experienced the difference between having the equipment and getting the job done.
A founder becomes his own customer
Cho’s technical training included a master’s degree and a doctorate in electrical engineering at Stanford. He completed the doctorate in 2013, then worked on Google’s speech team, training neural networks to transcribe speech. In 2018, he founded Bolt Communications to make speech-recognition errors easier to correct. The product concerned a modest but consequential question: once the machine makes a mistake, what can the person do next?
Marketing Bolt exposed him to another set of inconvenient next steps. He would later describe having to learn and juggle different tools as frustrating and inefficient. Getting a service in front of users required a succession of unfamiliar tasks. The technical founder had become a customer in a market whose complexity he could feel before he could explain it.
There is something useful about this kind of apprenticeship. A demonstration usually follows the path its maker has prepared. A customer has to deal with the turns after it ends. Cho’s eventual interest in marketing software grew from having to make those turns himself. By the time generative AI advanced in 2022, he had a problem in mind that could give the technology a job.
Three people, different kinds of fluency
Tofu’s founding team combined technical experience with someone who knew the marketing work from inside the department. Liu brought a background in machine learning. Zelby had spent eight years building and leading marketing teams at Slack, Consensys and Capriza. Cho brought engineering experience and the particular education of trying to promote a company himself.
He recalled a telling moment when he met Zelby: “She gave me my own pitch. I didn’t even have to say it.” The recognition was useful. A technical founder can describe a process that ought to improve; an experienced marketer can recognize the process from a Tuesday afternoon.
Before the launch, the team spoke with more than 100 enterprise chief marketing officers. The bottleneck described in its launch announcement was a shortage of resources for creating personalized content. Software alone had not removed that constraint. Teams still needed enough people and time to make material fit different buyers. The conversations gave the founders a specific place to begin.
The campaign has to survive the handoff
Consider an ordinary campaign. An email sends a prospective customer to a landing page. Both pieces need to explain the offer in a way that makes sense to that customer. If they are prepared separately, their language can drift. One promises a practical answer; the other seems to have arrived from a different meeting.
Tofu’s early approach was to create a Playbook from a customer’s website and existing materials. That shared context informed variations of landing pages, emails, whitepapers and other content. The same company information could travel into different assets. The reasoning is straightforward: a team should be able to adapt its message without explaining the company afresh for every piece.
There is a substantial difference between producing a paragraph and preparing material that belongs in a campaign. The paragraph needs words. The campaign needs a relationship among the words, the audience and the destination. Cho’s product sits in that relationship. Its usefulness depends on how much of the coordination it can carry for the team.
- 01Company materialsExisting message and brand
- 02Shared contextThe Playbook informs the work
- 03Campaign workflowAudience and channel decisions
- 04Connected assetsEmail · landing page · social
Illustration of the approach, not a product interface.
A webinar provides a more concrete example. In his February 2025 company update, Cho described a workflow that takes a recording and produces blog posts, social content and emails, then personalizes the emails for attendees. The event has ended, but the work of following up has only begun. Anyone who has organized an event will recognize this inconvenient scheduling arrangement.
He also described combining private CRM data with public material such as company news and earnings reports for account-specific campaigns. That moves the problem beyond writing in a chosen tone. The software has to connect information about the business doing the selling with information about the business being approached. Relevance requires both sides of the conversation.
Twelve times, with a date attached
Tofu announced a $5 million seed round led by Index Ventures in October 2023. In February 2025, it announced a further $12 million in Series A funding, led by SignalFire with participation from HubSpot Ventures and others. Across those two announcements, the company had raised $17 million.
The Series A arrived with company-reported figures: revenue had increased twelvefold over the preceding year, while content generated on the platform had increased thirty-sixfold. Named customers included Check Point, RingCentral and Bluecore. The figures describe a period when the product was finding a larger place in marketing teams’ work.
Funding announcements and company-reported revenue growth, as of February 13, 2025.
Revenue growth and content volume answer different questions. The first concerns the business; the second concerns activity on the platform. A marketer still has to care about what the material accomplishes. It is possible to admire a shorter production queue while wanting to know whether the resulting campaign helps a customer understand the offer. Cho’s argument for connecting the work ultimately has to meet that test.
The preparation behind the prompt
Cho’s public writing makes room for that human responsibility. In a LinkedIn post about the stigma attached to AI content, he pushed back on the idea that its users were simply producing generic material with little thought. “The best marketers using AI aren’t lazy,” he wrote.
“The best marketers using AI aren’t lazy.”EJ Cho
His examples involved preparation: investing in useful prompts, keeping templates for recurring work and supplying style guidance in advance. He described different ways of dividing the effort. Some marketers use AI to get much of a draft together before refining it. Others do most of the writing themselves, then bring AI into the remaining work. The timing depends on the task.
That observation gives his pitch a practical limit. Automation still needs decisions about what is worth saying and who should hear it. A reusable template preserves those decisions only if someone has made them well. The machine can help a team avoid repeating itself internally; the team remains responsible for having something useful to say externally.
His November 2024 conversation with Gaurav Kotak explored a similar set of concerns: personalized B2B content, the quality of AI output and the balance between automation and control. These are the questions a founder of marketing software has to live with. More output is easy to describe. Work that remains appropriate to its audience is a more exacting proposition.
Back to the space between
Cho has also taken the argument into public talks. His April 2025 AI User Conference listing paired a session on autonomous campaigns with a workshop on landing pages tailored to customer segments. His HubSpot speaker profile lists a talk about why fragmented stacks make life difficult for AI agents. The vocabulary keeps changing; the concern with disconnected work remains recognizable.
Tofu’s website now describes campaigns operating inside a company’s existing go-to-market stack. That is an instructive detail. A marketing team already has systems, information and habits. Any new product must find a place among them. Cho’s ambition to simplify campaign work therefore includes the unshowy business of fitting into the software people already use.
His route into the problem gives the story its continuity. Speech recognition raised the question of what happens after an error. Entrepreneurship raised the question of what happens after a product exists. Marketing raised the question of what happens after the draft is written. At each point, there was another task waiting for a person.
Cho has chosen to build around those tasks. The appeal of Tofu lies in a familiar wish: finish the campaign without becoming the courier between all its parts. For a founder who once had to learn that courier’s route himself, it is a sensible place to spend his attention. There is plenty of work between the tools.