$10 million and zero employees: Martin Varsavsky is building an AI doctor with an AI co-founder

$10 million and zero employees: Martin Varsavsky is building an AI doctor with an AI co-founder

19 November 2026

Martin Varsavsky has co-founded five unicorns over 30 years, in telecoms, renewable energy, fertility and now medical AI. His latest company, Certuma in Austin, aims to become the first AI doctor approved by the US FDA. When he raised a $10 million seed round from 8VC, the company had zero employees. It was him and an AI.

It's shocking that my co-founder can be an AI. Even in my own career, I had never seen a moment like this one to start companies.

This article condenses his interview on the Been There Done That podcast into the parts you can use: what stays the same in every company he builds, why he steps aside once a company starts to grow, and why so many of his companies were illegal when they started.

No playbook, just a way of building

Varsavsky is clear that he doesn't have a manual called Build a Unicorn. He also puts it in perspective: five unicorns over 30 years, not over a few months. Two companies are worth hundreds of millions, and two failed.

The most surprising part is that success still surprises him.

The daily life of an entrepreneur is very messy. What I see when I build companies is a lot of problems. The success is when you finally get your exit and you're just shocked that all that mess led to a successful exit.

He sees himself as an entrepreneur, not a vertical entrepreneur. When he got good at telecoms he didn't stay there, and he didn't stay in renewable energy either. That makes every new company virgin territory where domain expertise doesn't help much. What carries over is the general way of building a company. The downside is that people like to pigeonhole him: "but you're a renewable energy entrepreneur".

He sticks to two rules:

  • Every company is the last one. He always builds as if he will never start another.
  • Never CEO of two companies at once. He can chair three and run one, but the passion and intensity don't stretch to two CEO roles. Elon Musk is the exception that proves the rule.

Step aside when it starts to scale

This is where he differs most from Musk. Varsavsky runs the company himself until it has product–market fit, growing revenue and real users. Then he finds a COO who becomes CEO, and he becomes chairman.

They are the real leaders of these companies after I've taken them through the beginning. They are better than me at scaling these companies up, so I move aside.

It hasn't always worked, and sometimes he has had to search again. But he describes it as a consequence of never seeing a company as the only one he'll ever have: you know when it's time to bring in a better day-to-day operator.

About Musk, whom he knows, he says it's hard to learn from him. It's like going to an incredible concert and watching a piano player: do you play better piano afterwards? Musk has no assistant and no close team, never invests in other people's companies and has no co-founders. When Varsavsky discussed his fertility companies with him, the message was clear: if Musk got involved he would be in charge, and he would rather buy them. Two different methods that both work, and the point is not to copy the one that doesn't fit you.

A seed round with zero employees

Varsavsky has always been a product manager turned CEO. He used to have engineers building demos and mockups while he sketched by hand or in Figma. Two years ago he tried building with AI and found it wasn't good enough. Over the last six months that changed: he now builds his own prototypes.

That's how he raised $10 million without a technical co-founder and without a single employee. The cost was roughly $8,000 a month in tokens, which he calls "pretty affordable". The investors were fine with it.

After the round he hired a real team, including a CTO and a CFO. But AI stays in how he leads:

  • Agents read all his email, Slack and WhatsApp and summarize them.
  • Each company has its own knowledge base he queries when someone asks him something, and he can answer with a precision he never had before.
  • The agents are starting to understand what matters to him, and even the mood in the company: the relationships between managers and how optimistic people are.

You tend to think that you read your emails. And you probably do. But you never read your emails like an agent can read your emails.

He flags the trap himself: being able to remember everything doesn't mean you should get stuck in the past. The point is to surface what matters most.

The idea came from a stye

Certuma started with his own problem. Varsavsky got a stye, photographed his eye and asked an AI. The answer was right: a stye, and a prescription drug. At the pharmacy in Miami he was told he needed a prescription. So he asked the AI for one, and it replied: I'm not a doctor.

How can it be that the AI tells you everything you have, looks at the photo, gives you a medication, and then when you ask for the prescription, it says I'm not a doctor? Somebody's going to do it. I'm going to do it.

So the problem he is solving isn't that AI can diagnose. That's already obvious. The problem is that AI isn't allowed to prescribe, and solving that means marrying technology and regulation. In his first comparison against telehealth, Certuma was right 98.7% of the time and the doctors 96%.

With AI he could also map every possible route to approval himself: how the FDA regulates devices and treatments while the states regulate the practice of medicine, and what the politics look like. That used to take books, articles and experts.

Many of his companies were illegal when they started

This is the thread he highlights himself. When he started telecom operators, the industry was a monopoly and it was illegal to be a new operator. He had to lobby and persuade governments. The same happened in renewable energy and fertility.

He describes healthcare today as Napster before Spotify: a kind of medical piracy where the AI already tells you what you have but isn't allowed to take the last step. Regulators see the value but worry, rightly, about hallucinations.

The lesson from Spotify and Netflix is that making something legal isn't enough. You also have to make it better. BitTorrent required hard drives full of files. Spotify made it easier and paid the rights holders. Certuma's equivalent is that the language model doesn't stand alone:

We use LLMs, but they're corroborated by deterministic software that we developed. That is our unique component, not the LLM, which anybody can get.

The deterministic layer makes sure the AI never deviates from the standard of care, never prescribes something that can't be prescribed, and never gets the drug or the dosage wrong.

Right vision, wrong execution

Two of his companies didn't do well. In both cases he was right about the trend but wrong about the execution.

  • Einstein Net was one of Europe's first cloud computing companies, around 2003–2005. Everyone held on to their laptops, and Varsavsky lost about $100 million of his own money before selling for almost nothing. The buyers then rode the cloud wave.
  • Fon became the world's largest Wi-Fi network and was worth hundreds of millions. But he didn't foresee how good mobile would get, and when the EU abolished roaming charges much of the point disappeared.

I haven't been wrong on the tech. I haven't been wrong on the vision. But I've been wrong two times with the execution. I didn't see other trends that would make my product less useful.

Be open about the risks

Certuma's biggest risk is regulation: what if medical AI isn't regulated at all, or it takes too long? He says so plainly, to investors and in the interview. He also points to the political headwind against AI, which isn't helped by the AI companies' own leaders warning every day that the technology might kill us.

But the risk is also the deal. Invest once the regulatory path is clear, and the valuation is higher and the return lower.

He has taken companies public, and a prospectus is full of risks. That trained him to disclose them. Investors are sophisticated, and if they see a founder unaware of risk they may pass. Things often go wrong in predictable ways, and then it's because you didn't mitigate enough.

Location is an opportunity

Certuma is being built in Austin. Varsavsky lives in Miami for family reasons, but for day-to-day building he thinks Austin is better. He used to build more in New York and California, but there employees live worse lives on the same salary, because taxes and rent eat more of it. The same goes for the big payout at an exit.

For founders in Europe, he sees London as the place with all three things you need: great universities, VCs and founders. Other cities have parts of it, but not all three. Extraordinary companies can happen in unusual places, but they generally don't.

Three takeaways

  1. Build the generalist's muscle, not the domain's. Varsavsky changes industry every time. What carries over isn't industry knowledge but the ability to build a company and to see what's new as it is. He calls it the fine line between experience and prejudice.
  2. Look for the gap between what the world already believes and what the rules say. Telecoms, Spotify, Uber and now medical AI: society is already convinced something should be legal, but it isn't yet. Whoever makes it legal, and better, wins.
  3. AI has moved the starting line. An experienced founder raised $10 million with AI as his only co-founder. The prototype, the regulatory research and the overview of the company could all be done alone. For anyone starting now, the barriers have never been lower.

Based on an interview with Martin Varsavsky, founder of Certuma, on the Been There Done That podcast. Figures are those he gave at the time of the interview.