Vasek Mlejnsky

Vasek Mlejnsky grew up in a Czech town of about 20,000 people, the kind of place where the wider technology world arrives only through a screen. He found that world early. Alongside a classmate he had known since sixth grade, Tomas Valenta, he started building video games. The two shared a curiosity that outgrew their surroundings, and somewhere along the way Mlejnsky came across Paul Graham's essays on startups. Reading them changed the question he was asking. Coding stopped being only about games and became about what else software could do.

That shift pointed the pair toward developer tools. They experimented with several ideas - optimizing serverless functions, a search engine for developers - looking for the thing they actually wanted to work on for years. In 2021, Mlejnsky made the move that most people talk about and few make: he left the Czech Republic for San Francisco, arriving through the Pioneer accelerator program. His reasoning was practical. For an AI-focused developer tools company, he believed proximity to users mattered more than anything. "Being in Silicon Valley is crucial," he has said. "It's about meeting your users, having hacking sessions." He puts it more bluntly elsewhere: "I was born European or Czech, but I belong to San Francisco."

The idea for E2B came from a frustration Mlejnsky felt while building his own agentic system. He kept running into the same wall: the cloud, as it exists, was designed for humans. People log in, click, type, and manage servers. An AI agent needs to do all of that too, but the tooling around it assumed a person on the other end. If agents were going to write code, browse the web, analyze data, and build applications on their own, they needed somewhere to actually do it - an isolated, secure, disposable computer they could spin up in seconds and throw away when finished.

That became E2B, which he co-founded with Valenta in 2023. The product is an open-source cloud runtime that hands each AI agent its own sandbox. Under the hood, those sandboxes run on Firecracker microVMs, the same lightweight virtualization technology that powers AWS Lambda. Each one is isolated from the others, which matters enormously when the code being run was written by a machine rather than a vetted engineer. Agents get a real file system, a browser, a code interpreter, and persistent storage, all wrapped in security boundaries an enterprise can trust.

Mlejnsky is deliberate about how he describes it. He calls E2B "a boring AI code execution layer that should just work." That word - boring - is a choice. In a market saturated with claims about intelligence and autonomy, he has positioned his company around virtualization, sandboxing, and security. Not the model. Not the framework. The plumbing. "Enterprises have massive expectations for AI agents due to all the hype around them," he has said, "but they won't easily scale if built on legacy infrastructure not designed for them."

Mlejnsky's bet is a big one. He thinks AI agents will become "as commonplace as apps on your iPhone" within a decade, and that each of those agents will need its own secure, dedicated computing environment - the same way every iPhone app leans on iOS beneath it. If that future arrives, the number of running sandboxes climbs into the billions, and the infrastructure to manage them has to be built now.

He frames the long-term goal as "an automated AWS 2.0, where AI software can deploy infrastructure independently." Today a human provisions servers. In his version of the future, the software provisions itself. Speed is central to why this is hard. "All main use cases for AI agents need extremely high speed," he has said. "Agents building websites and research reports need to answer in low seconds." A sandbox that takes a minute to start is useless to an agent expected to respond instantly.

Open source is part of the strategy, not a side effect. E2B grew from a public project that collected thousands of GitHub stars within weeks of launch, and the company continues to invest in interoperability with other infrastructure providers rather than locking developers in. It is a wedge and a philosophy at once: standardize the environment that agents run in, and let it work across clouds.