Genspark COO Wen on Building an AI-Native Unicorn

Genspark's first $10 million in ARR took nine days.

That number came up almost in passing when we sat down with Wen, co-founder and COO of Genspark, on The QAI Podcast. The company went from zero to $50 million ARR in five months, has grown 20–30% every month since, and crossed unicorn status in under 20 months. The nine-days line is the one that sticks, and Wen is candid about why it happened: when Genspark sunset its AI search product, it still had a few million users and a newsletter list. "We built a good product that people loved, and then we just push it out on LinkedIn, X posts, Instagram." Word of mouth did the rest.

The growth curve isn't the interesting part, though. How a small team builds this fast is — and what Wen thinks happens to knowledge work once your tools start acting like teammates.

"We don't have to make decisions, we just build it all"

Most startups with 150-plus tools and a horizontal product carry a feature-team org chart to match. Genspark doesn't. There's one product team, and every engineer is full-stack — not just on engineering, but on product too. "Each one or two engineers would be leading one complete product behind the scenes," Wen said.

That works because of how the company is wired. Wen calls Genspark an AI-native company, and gives it a concrete definition: "80%-plus of the code is written by AI." The team sits on top of 30-plus SOTA models, 150-plus in-house tools, and a data layer with 20-plus premium databases that keeps outputs grounded. The result is a different relationship with prioritization. "We don't have to. We build everything we want," he said. "We just build it all."

There's an obvious objection, and Wen doesn't dodge it: if AI writes the design, the tests, and most of the code, how does anyone hold the quality bar? Part of the answer is structural. A few co-founders did the hard infrastructure work first. Justin spent 12 years at Facebook, scaled Messenger from zero to billions of users, and was a founding member of PyTorch. Another co-founder was early at Google on the search-ranking platform; another, Lenjoy, built teams at Pinterest, YouTube, and TikTok. "You have to have a group of awesome architects to build that scalable technology foundation," Wen said. "And then you provide that platform to every engineer on the team, and then they can just go nuts."

The second half of his answer is the part that lands. Genspark runs rigorous testing, much of it AI-written — but the philosophy underneath it is human. "AI would do the work for you, they report to you, but they don't take responsibilities for you." He applies it to himself: he doesn't write his own emails anymore, he has Genspark scan his inbox and draft responses. He still proofreads every one. "Because once I click that send button, that's me, not my AI."

Agents as teammates, not job-killers

Wen has heard the "AI is going to take everyone's jobs" line, and he answers it with a homely analogy: nobody mourns the job the washing machine took. His framing is that Genspark serves the billion-plus knowledge workers who, by his estimate, spend 80% of their time on busy work — fixing a logo on a slide, repairing one formula in one cell of a spreadsheet. Strip that away, he argues, and you elevate people to "the creative, strategic real work."

He doesn't wave the disruption off, though, and this is where the conversation gets more honest than most founder pitches. "I'm not saying we should just ignore the possible risks, challenges, and even crises," he said, naming wealth distribution and job losses directly. His own read: junior roles are getting cut faster than new ones are created, and "human beings compete with human beings, human beings don't compete with machines." AI, in his view, "shouldn't be something that only benefits a small group of people."

His prescription is concrete. He told the story of Matthew, a college grad in Michigan who builds Shopify sites for local merchants at $6,000–$7,000 a pop. Matthew used to hire offshore developers on Upwork for two or three thousand dollars a project. Now he pays about $25 and pockets the rest — doubling his margin, serving three or four times as many clients, with more time freed to think and scale. The lesson Wen draws: "Don't just test AI, actually use AI to let AI benefit you." His blunter version, for anyone betting AI won't matter in five years: "Don't be the ostrich putting your head into the sand."

From AI search to a "super agent"

Genspark didn't start with agents. It started with AI search, at a moment when, as Wen put it, "for 20 years there's no one would ever doubt or question Google, ever." The team generated multimodal Spark Pages instead of link lists and hit roughly 5 million users fast. Then came the problem every AI-search company hit: how do you make money off it?

Two things unlocked the shift to a general agent. One was a product question they kept asking from the user's side — people don't want information for its own sake, they want the job done: "why do you stop at just serving the information?" The other was timing. Before 2025, Wen noted, LLM reasoning was limited; once models like Sonnet 3.7 and OpenAI's o-series made planning and reasoning practical, the agent became viable. As he put it, LLMs "are like brains, they don't have arms and legs to do the real work. You have to build arms and legs" — those are the 150-plus in-house tools and the context engineering around them. Genspark launched its super agent in April, and Wen says they've shipped a "world's first mixture of agents" architecture that lets them switch on new models as they arrive. They now get early access to test models for OpenAI, Anthropic, and Google before release.

On November 20th the company launched Genspark for Business, with team and enterprise plans, after a summer of inbound from PE firms in Japan, banks in London, oil and gas companies in Bogotá, government agencies in Dubai. That led to the partnership the episode opened on: Genspark is launching on Microsoft's Agent365, and Wen had just met Satya Nadella at Microsoft headquarters. By his own accounting, Genspark has executed only "10, 15% of product roadmap" so far.

The takeaway

What Wen kept circling back to is a working definition of human value in an agent-heavy world: not hours logged, but results owned. "We are gonna be valued now based off results we deliver — your taste, your judgment calls, your insights." That's a useful frame for any founder building with agents right now. Let the model do the work and report back. Keep the send button.

It's a thesis we recognize at QAI Ventures, where the bet is that customers, code, and distribution — not just capital — are what carry AI-native founders. Genspark, building most of its own code with AI and growing on word of mouth, is a clean example of what that looks like in practice.

Watch the full conversation with Wen on The QAI Podcast or browse more episodes at /podcast.