Vasek Mlejnsky is co-founder and chief executive of E2B, an open-source infrastructure company he founded with Tomas Valenta to give AI agents secure cloud computers. His defining contribution is making secure code execution for AI agents practical: isolated environments where models can write software, manipulate files, analyze data, and return usable results.
Mlejnsky studied at Charles University and built early software projects with Valenta, whom he had known since secondary school. Their company, Devbook, initially offered programmers a desktop search tool for Stack Overflow and GitHub; Mlejnsky introduced its first version in December 2020.
By April 2022, Devbook had become an SDK for embedding executable examples in developer documentation. Its per-user virtual machines already incorporated Firecracker isolation, ephemeral environments, Dockerfile-defined configurations, and machine snapshots. In April 2023, Mlejnsky and Valenta adapted that infrastructure into E2B, redirecting secure computing environments from human developers toward autonomous software agents.
In July 2025, Mlejnsky announced E2B’s $21 million Series A to expand enterprise deployment and its open-source ecosystem. More recent applications include AI-generated software development and financial research, where agents need executable environments, controlled networking, and auditable operations.
- An agent needs a usable computer. Mlejnsky treats the sandbox as a complete Linux environment capable of running terminals, installing dependencies, launching development servers, and producing files or visualizations. Code interpretation is one application of that infrastructure, not its boundary.
- Firecracker microVM isolation preserves capability. Separate virtual machines isolate users and workloads while giving agents ordinary operating-system access. Dockerfiles specify an environment’s dependencies and filesystem; E2B converts that filesystem into a Firecracker virtual machine instead of treating the production sandbox as a conventional container.
- Persistent sandbox sessions make agents iterative. In his AI Engineer World’s Fair workshop, Mlejnsky built a Claude-powered application that reconnects each user to a Jupyter-backed environment. Variables, files, and previous computations survive successive requests, while execution errors and generated charts return to the model or interface.
- Snapshot-based agent branching enables exploration. Saving a virtual machine’s filesystem and memory would allow an agent to test alternative actions, restore earlier states, and explore multiple execution paths without rebuilding its environment.
- Production autonomy requires operational control. Mlejnsky emphasizes session-level isolation, network allowlists, auditability, and sandbox lifecycle observability. His public update on resource monitoring highlights visibility into resource consumption and pause-and-resume events as agents take on longer-running work.
