Deep Level.
The Founder Magazine
Founder & CEO Interviews
Est. 2024
Independent & Reader-supported
Subscribe
Interview 7 min read

Two Years of Nights and Weekends Before He Quit

They want accomplishments. They want to complete something in their career. That's why they join your company.

Zheng Li
Zheng Li
Co-founder & CTO, Myko AI
The Company, in Brief
Sector
AI Software

Zheng Li did not leap. He spent two years working on someone else’s idea in the evenings and at weekends while holding down a full-time job at another startup, and only then did he commit.

What is instructive is his account of what finally made the decision possible, because it is not a story about belief. It is a story about three separate conditions arriving at once.

“After two years, my wife got a job, so financially I don’t have such a big burden,” he says. “And also, we had some traction.”

Then the third one, which founders rarely name out loud.

“And also, after two years, I know this person better. So I think I can work full time with this person.”

He went full time in October 2022. The two years were not hesitation. They were due diligence, on the market and on his co-founder simultaneously, paid for by someone else’s payroll.

The whole stack, from the bottom

Li moved to the United States from China at 21, having already finished a degree there. What followed was an unusually wide education: materials science, then computer science, then an MBA at Columbia.

His career began at the very bottom of the technology stack, as a compiler engineer at Nvidia, back when the company was known mainly to people who played video games.

“I started very low in the stack, very close to hardware,” he says. “It’s like building a house. I really understand the foundation.”

From there he moved steadily upward: a startup building virtual drivers, then model work, then perception models for self-driving cars, and now a product used directly by consumers. He was employee number two at one startup and its first core technology hire; it raised a Series A and was acquired by VMware.

Most engineers spend a career in one or two layers of that stack. He has worked in all of them, in order, which is why he can reason about a consumer AI product in terms of what the GPUs underneath are actually doing.

He also had a startup of his own before this one, and its ending is a lesson in itself. He used machine learning to cut basketball highlights and upload them to YouTube. It worked. It could not be monetised.

“All the copyrights belong to the NBA.”

That failure had an unexpected payoff. It made him known in his business school’s community as the technical founder, which is how his eventual co-founder found him: with a programming question he could not solve on his own.

Hiring on two point seven million

Myko raised $2.7 million, and Li treats that number as the defining constraint on how the company is built.

“We have a lot of restrictions in terms of hiring. We don’t have that much money to spend, as many other startups. That’s why we need to be very careful of hiring the best talent.”

His response was to throw out the standard technical interview.

“We don’t use LeetCode. We have our own tests. We have a take home project. It’s open book. Everything is open.”

The project is real work: build a small application from scratch that solves a problem the company actually has. It runs up to a week, with the candidate sending intermediate results and the team giving feedback along the way.

What that buys him is worth more than a puzzle score, and he lists it precisely.

“One, we understand that he can think. Two, we already have this experience of working together. And three, I know how to manage this person.”

By the time an offer goes out, the working relationship already exists. The interview was the first week of the job.

People do not join for the rate

Li’s view of what keeps good engineers is blunt, and it is aimed squarely at founders who think compensation is the whole answer.

“It’s not like, to earn how much dollar per hour. That’s not their goal.”

What they want instead is completion.

“They want accomplishments. They want to complete something in their career. That’s why they join your company.”

Which turns motivation into a design problem for the founder rather than a budget problem. “You have to think about challenging problems to give to them. And that problem also fits into your company’s overall goal.”

The operating rhythm underneath is unremarkable and consistent: a standup every morning, a one-to-one every week. He is clear that the one-to-one is not a status report.

“You know this person, how this person thinks, how much time this person spent on this, how much this person knows about this problem, by just asking questions. And also in this process, you also help this person think through.”

Where the line between contractors and staff sits

With a small raise, Myko runs a hybrid team: a small full-time core plus contractors hired through Upwork. Li is unusually specific about where the boundary falls and why.

Contractors buy flexibility. “Even if you are not happy with the person, just let that person go.” The trade-off is availability, since many people contract precisely because they want flexible hours, so “you cannot rely on them 100%”.

His practical fix is to hire hourly rather than fixed price, then agree an expectation each week. “You can talk to them, like, your expectation is 30 hours a week, 40 hours a week. If the guy is really good, then you can push to 40. So that person is like a full time, but it’s still flexible.”

The line itself is not about seniority or cost. It is about intellectual property.

“It’s also an IP problem. You don’t want to expose your core IP to the contractors.”

So the core intellectual property stays with full-time employees only, and the full-time bar is correspondingly high. “For the full time, they have to be really good in some area. That’s what you need the most.”

It is a coherent answer to a question most early-stage founders resolve by accident.

Designing against hallucination

Ask Li what is actually wrong with current AI and he does not hedge.

“The biggest problem with AI, especially the language model, is the hallucination problem. It cannot find information, so it just hallucinates.”

His criticism of the mainstream approach is architectural rather than dismissive. General assistants answer immediately, from memory. “It does not try to avoid it, because you ask something, it responds immediately. It does not really think through.” And memory is the wrong source. “It searches from its memory first, because it memorises a lot of things, and the memory sometimes is wrong.”

Myko’s design inverts the order. Search first, answer second. Then a second model audits the result before anything reaches the user.

“We have a reviewer, a reviewer language model to review every answer, every single step. We try to examine any intermediate result.”

If the reviewer is not satisfied, the system searches again. He is careful not to overclaim. “We cannot get rid of it totally. But you can try to avoid it.”

The underlying principle is one any technical founder can apply. Reliability comes from how you arrange the system, not from asking it more nicely.

Honesty, and playing defence

Asked about values, Li reaches first for football. He is a Manchester United supporter since childhood, and the players he admires are defensive midfielders who cover the whole pitch.

“If I join a team, I usually am the one who plays the hardest defence,” he says. “But sometimes we see people just kind of walk around, don’t play defence.”

The second value came from his first employer, and he still remembers the format.

“When I joined Nvidia, there was a small notebook. It talks about the company values. One thing I still remember is honesty. You have to be honest to your peers, you have to be honest to your managers.” He means something concrete rather than aspirational. “If there’s some issue with the system, some bugs, you cannot hide the bugs.”

The third is a quality he actively wants pointed back at himself.

“They don’t have to usually follow instructions. They should also critically challenge your instructions, challenge your direction.”

For a CTO hiring on a tight budget, that is a deliberate choice. The people he can afford are few, so each one has to be capable of telling him he is wrong.

Key Takeaways
  • № 01
    Do not quit on conviction alone. Zheng waited until traction, his finances and his trust in his co-founder all lined up, which took two years of evenings and weekends.
  • № 02
    Replace the whiteboard puzzle with a real, open-book project on a problem you actually have. By the time you make an offer you have already worked together for a week.
  • № 03
    Engineers do not join for the hourly rate. They join to finish something worth finishing, so the problem you hand them has to connect to the company goal.
  • № 04
    Draw the contractor line at intellectual property rather than at seniority or cost. Core IP stays with full-time staff only.
  • № 05
    Hallucination is an architecture problem. Myko searches before it answers, then has a second model review every intermediate step.
About the guest

Zheng Li

Zheng Li, who goes by Lee, is co-founder and CTO of Myko AI, which builds an AI assistant for research and everyday digital tasks. He moved to the United States from China at 21, studied materials science then computer science, and later took an MBA at Columbia. He began his career as a compiler engineer at Nvidia and has been an early employee at two startups, the first of which was acquired by VMware.

← Back to the Index