The Last Mile Is the Hardest
EJ Cho’s decade-long journey turning early AI into the future of marketing execution
Written by: Alex Sventeckis
Summary
- Cho spent 10 years preparing for a future nobody else could see: Training neural networks at Google in 2013 and building failed startups gave him the exact skills needed when ChatGPT finally made the market ready for his vision in 2022
- He rejected the "GPT wrapper" gold rush to solve marketing's hardest problem: While others built 50 identical AI SDR tools, Tofu tackled true workflow orchestration, delivering finished assets (Google Slides, emails, landing pages) not just suggestions
- Product-market fit wasn't found, it was molded through constant rejection: Cho and cofounder Liu spent years "constantly ideating and launching stuff every week," speaking with 40+ CMOs, and adjusting until they shaped something that stuck
- Being an outsider became Tofu's superpower in a frozen industry: Without marketing backgrounds, the founding team questioned everything established players accepted as "the way it's always been," leading to $17M in funding
- Cho's philosophy: "A lot of pivots are just running away from hard problems": Like companies Clay and Ramp, Tofu persevered through the painful journey until the market caught up, proving conviction beats chasing easy wins
Introduction
In 2013, Eunjoon (EJ) Cho had just completed his PhD in electrical engineering at Stanford and was training neural networks at Google to transcribe human speech. A decade later, the same technology would transform marketing.
Back then, he simply wanted to make machines listen better. But his early AI work would prepare him to venture into the unmapped forest of generative AI. Doubt stalked him, as it does any founder shaping nascent technology into something revolutionary. The world kept trying to get him to pivot — to quit on his vision.
Cho and his marketing workflow automation startup, Tofu, are a testament to smart persistence.
Accidentally Preparing for the Future
It feels impossible today that the term “large language model” wasn’t ubiquitous. But there was a time when every tech journal article or LinkedIn post didn’t mention AI, LLMs, or ChatGPT.
Cho spent those early days at Google, training and launching language models as part of the speech team. They weren’t called LLMs, but these models formed part of the foundation for what we use today. Cho glimpsed what these models — and, by extension, AI — could one day create.
He also saw glimpses of a future outside of Google. Cho distinctly remembers his third year at Google, seeing CEO Sundar Pichai walk by him and sparking an epiphany:
Cho did not want to be Pichai.
Climbing the ladder wasn’t Cho’s path forward in life. He loves ambition and risk, a fan of scrappy startups, not comfortable behemoths.
“I call it an early midlife crisis,” he said. “I was doing well at Google, but I looked up and thought that even if I got promoted all the way up, it didn’t feel that appealing.”
His experience told him something was amiss with speech recognition. Input error rate was low — around 5%, or one out of twenty spoken words. But when you catch an error with Alexa or Siri, there’s no backspace button to fix it. It’s a “brain fart moment” that confuses users, as Cho notes. “You don’t know what to do next.”
Cho founded Bolt Communications in 2018 to build an intuitive correction interface to help speech recognition users fix brain farts — a vocal backspace button. It had tremendous potential for users reliant on voice input, like older adults or people with hand tremors. But that market proved too narrow, and traction stalled.
While the startup didn’t take off, it inspired what Tofu would become. During Bolt, Cho spent significant time on its B2B go-to-market efforts. He wanted a solid marketing tech stack to accomplish his goals, and he explored tool after tool…after tool…that promised to help.
“It’s a very fragmented space. For everything, there’s a separate way and tool for doing it,” he said. “Let’s say I want to do some organic growth; I need to figure out SEO. I had no clue. Email? I need a tool for that. Website? Another tool. Every paid channel is different.”
Founders often hear that failure is a wise teacher. In Cho’s case, failure also painted the path forward.
“Not all smart people at large companies would take risks. And I wanted to take risks.”
— EJ Cho, CEO at Tofu
Not All Who Wander the AI Wilderness Are Lost
After Bolt, Cho returned to Big Tech with an engineering role at Meta (then Facebook). In the Bay Area, moving between startups and Big Tech is common. “You can leave Big Tech, try a startup,” said Cho. “If it doesn’t work, you can come back; it’s not a big deal.”
Cho sensed that the attitude of risk-taking would soon be ripe for AI, but it wasn’t there yet. He joined Meta around the COVID-19 pandemic, when the company was infusing conversational AI and multimodal experiences across its platform. More work with AI reminded Cho of generative AI’s budding promise.
Nobody reminded him better than Honglei Liu, who would become one of Tofu’s cofounders. Bolt had taught Cho that he was a generalist founder, strong with product but needing deeper support with engineering execution. In comes Liu, “a machine who can execute.”
Putting Liu and Cho together would prove a wise decision as they wandered the startup wilderness together. After two years at Meta, Cho knew the right startup for him waited somewhere in the dark forest of early AI.
There’s a thrill in getting lost in the heady act of pure exploration. Though he held a few other roles before founding Tofu, Cho spent his free time with Liu mapping the generative AI market. They’d forge new ideas, test them with customers, and drop them just as fast: “We were constantly ideating and launching stuff. Constantly. Like, every week.”
That’s the trouble with seeing the future before everyone else. The generative AI market was nascent, still immature. AI riding on the backs of Google and Meta? Sure, maybe something was there. But the market couldn’t support wholly new AI-first companies.
That didn’t stop Cho. Exploration burned within him, and he kept eagerly investigating new AI terrain. “Not all smart people at large companies would take risks,” he said. “And I wanted to take risks.”
The ChatGPT Moment: Timing and Ambition Align
The sun was just rising over a quiet patch of Siberian forest on June 30, 1908, when the sky exploded. A 130-foot-wide asteroid detonated six miles above ground, flattening 80 million trees. The Tunguska Event awakened a sleepy forest with a mighty crash.
In November 2022, ChatGPT exploded onto the tech scene and awakened the AI forest.
For many (myself included), ChatGPT arrived seemingly from nowhere. But just like the Tunguska asteroid spent millennia tumbling through space until it hit Earth, ChatGPT’s underlying generative AI framework had been long in the making.
Technology finally caught up with Cho’s vision. Foundational models could handle actual work. Founders could build quickly using APIs. And investors and customers finally realized the value in adoption.
Cho could finally plant Tofu in fertile ground. He found product validation from his first meeting with Elaine Zelby, a VC-turned-operator who became Tofu’s third cofounder. A marketer by trade, she basically pitched Tofu’s value proposition back to Cho.
“She gave me my own pitch. I didn’t even have to say it,” he said. “I knew we had to work together.”
By 2023, Tofu had the right group of complimentary cofounders to explore a generative AI forest alive with possibility.
The Rewards for Solving Hard Problems
OpenAI literally opened AI and its capabilities to millions of users. Many used that power to go for easy wins.
If you perused Product Hunt in 2023, you witnessed a product graveyard in the making. Slick interfaces disguised what most tools actually were: clever wrappers around GPT API calls.
“There are fifty AI SDR tools all doing the same thing,” said Cho. “Everyone’s building demos or prototypes, but very few are focused on the actual workflows or delivery. That’s where the gap is.”
Not all AI companies are GPT wrappers, of course. Many use foundational models from OpenAI or Anthropic as backend workhorses while delivering real value elsewhere.
But the fragmented tool market that frustrated Cho in 2018 didn’t magically fix itself. If anything, it got worse.
Many startups wanted to solve surface-level problems because they’re easy and feel productive. But almost anyone with API credits can do it. You’re one of a thousand replaceable options.
Hard problems are messy, complicated, and unglamorous. Most people avoid them, but it’s where the real value lies.
Tofu is out to solve true marketing workflow orchestration. Integrating with CRMs like HubSpot, formatting assets, and adapting to campaign context are important but hard problems. But that’s what marketers want AI to handle: the “last mile” of turning ideas and drafts into finished, publishable, on-brand assets.
“The output has to be in the end format of that asset,” said Cho. “That’s what people miss when they think about AI. It can’t just be a suggestion — it has to be the thing you can post or ship.”
That’s what Tofu delivers. Not suggestions or drafts, but usable assets: completed Google Slides, ready-to-send emails, or publishable landing pages.
And unlike tools that treat prompts like one-offs, Tofu builds context across assets. Its orchestration capabilities help marketers manage full omnichannel campaigns with an AI tool that actually learns. AI upgrades from a writing assistant to a true marketing execution partner.
This is a hard problem to solve. But it’s the one that delivers on AI’s promise to marketers: good work done well across every channel. For comparison, a 1080p high-definition video at 30 frames per second on your iPhone requires 130 megabytes (MB) per minute. According to Cioni, professional video files can run one gigabyte (GB) per second. Digital storage also got better and cheaper. When Cioni started Plaster City in 2003, one terabyte (TB, or 1,000 GBs) of memory on disk cost $1,055. In 2023, that terabyte cost $11.
Everyone’s building demos or prototypes, but very few are focused on the actual workflows or delivery. That’s where the gap is.”
— EJ Cho, CEO at Tofu
Honing an Outsider's Edge
Cho has wandered unknown lands before. He grew up as a nomad, attending seven different elementary schools across continents, from London to Singapore.
The greatest lesson from that time? People are similar across cultures. We share motivations and frustrations. We care about the same things.
Cho’s global background honed his product intuition. He can connect with customers across contexts and adapt quickly when things change.
“Wherever you throw me, I think I can figure things out. That’s just how I’ve always felt,” he said. “I’ve lived in a lot of countries, moved around a lot, adapted to a lot of environments. I think that gave me a belief that I could handle uncertainty.”
In the ludicrously fast-moving AI market, uncertainty is a feature, not a bug. Handling it is a superpower. That mindset positions Cho and Tofu to challenge assumptions, question processes, and inject novelty in markets that desperately need it.
Because no matter your field, you’ll eventually hear, “This is the way it’s always been.” Whether it’s a clunky tool, an outdated process, or a fossilized way of thinking, time and inertia can freeze progress.
Marketing is no exception. Many tools teams use today simply exist because “it’s the way it’s always been.” Cho rejects that mentality, instead leaning into the advantages that fresh perspectives can bring.
He also knows how hard it can be for founders to feel like they belong. Industry loves its domain experts — and they certainly have value. But sometimes, a little domain adjacency shakes things up the right way.
Cho turns to the philosophy of DoorDash founder Tony Xu, who believes people can become effective operators in unfamiliar fields faster than they think. What holds them back? Never starting.
“Sometimes you feel out of place, especially if you’re not from the ‘in crowd,’” said Cho. “But you can just do stuff. You can just start building. That’s the only way to be taken seriously anyway.”
Tofu’s founding team embraced that philosophy from day one. They weren’t domain experts, so they needed to learn. The team started by speaking with over 40 B2B CMOs, building their knowledge of what really irked marketing leaders.
“We didn’t come in with a playbook. We weren’t marketing execs. So we had to talk to users constantly — not just to validate, but to understand,” he said. “I think that outsider lens actually helped us stay focused on what people actually needed, not what we assumed they wanted.”
That outsider focus worked. Tofu has raised around $17 million between seed and Series A funding. The seed round was a bet on the team’s potential. The Series A proved AI had crossed an inflection point. Others finally saw the future Cho envisioned.
One might look at Tofu’s success and say they stumbled into product-market fit — a lucky break in a hot category. But Cho’s years of building and iterating show that PMF isn’t found; it’s made.
“There was a lot of molding product-market fit in the early days. We were adjusting constantly, figuring out what stuck. It wasn’t like it just clicked — we had to shape it.”
Now, as the AI forest comes alive, Tofu’s challenge is to scale while keeping its edge.
“Speed matters more than you think,” he said. “You can’t just wait for the perfect product — you have to keep shipping, watching how people use it, and adapting fast.”
Tofu’s 20 employees share Cho’s ambitious, risk-taking mindset. It’s led to a company culture that embraces taking chances, breaking things, and learning from mistakes.
“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,” said Cho. “That’s how we move fast and learn fast.”
An Unknown Distance Yet to Run
In 1869, John Wesley Powell led an expedition to the American West, mapping hundreds of miles of rivers and exploring the Grand Canyon. His crew lost boats and braved unknown rapids along their way to success.
At the journey’s most uncertain point, Powell wrote:
"We have an unknown distance yet to run, an unknown river to explore. What falls there are, we know not… Ah, well! we may conjecture many things."
But what if Powell had given up instead? What if he’d surrendered before charting the Colorado River, savoring the majestic Grand Canyon soaring overhead?
Uncertainty about the future can frighten even the most stalwart startup. Just as Powell could’ve turned back, founders can pivot from hard problems and chase the sugar rush of easy wins. And many “AI-powered marketing startups” did exactly that.
But not Tofu. Cho saw marketing’s last-mile problem — hard but valuable, something AI could solve. Getting there meant spending years navigating through uncertainty. The wrecks of Cho’s failed ideas, false starts, and dead-end markets dot the riverbanks.
The rapids of an untested market are terrifying, but the river does calm in time. Still, many founders surrender before calmer waters arrive. They miss the majesty of success achieved through conviction and persistence.
Cho and Tofu have succeeded by combining technical expertise, market pain, and patience. They molded product-market fit into existence and were ready when waters calmed.
During moments of struggle, Cho could’ve pivoted to something else. But by his token, “a lot of pivots are just a way of running away from hard problems.”
He points to companies like Clay and Ramp that persevered through long, painful journeys until they hit their moment and skyrocketed. As he sees it, conviction carries companies through challenging times until the market is ready.
“You’ve got to work through it. There are always ups and downs, but that conviction and persistence is so important.”
Most of AI remains uncharted, and it’s evolving fast. Already, AI agents are executing more marketing tasks within more teams. One day, AI could even outsmart the teams it’s meant to help.
Regardless of AI’s future, Cho’s lesson holds: embed yourself in your customer’s world. Become an expert. Ford the river and ride out the rapids to reach calmer waters.
You may not know where the river ends. Venture forth anyway.
Author: Alex Sventeckis
Alex Sventeckis is the director of content strategy at Weaver Fundraising and an instructor of marketing at Ball State University’s Miller College of Business. With expertise in content strategy, creation, editing, and team management, he helps brand and communications teams build content into a revenue-generating function.
