Daytona raises $24M Series A to provide secure sandbox infrastructure for AI agents

Summary

Daytona, a three-year-old infrastructure company, raised $24M in Series A funding for its platform that provides isolated sandbox environments for AI agents to execute code, run browsers, and support reinforcement learning training.

Eight months ago, the company rebuilt its product from scratch and fired existing customers. Co-founder and CEO Ivan Burazin describes the shift as a recognition that agents operating like digital knowledge workers need actual computers rather than just API connections to external services. Real work requires downloading files, installing software, and running local analysis, tasks that demand a dedicated computing environment.

Daytona supports three use cases: code execution, where products like Lovable spin up sandboxes for every user action; browser automation for QA testing of websites; and reinforcement learning environments where researchers need to spin up hundreds of thousands of concurrent sandboxes within minutes.

Since the product launch eight months ago, Daytona has added customers ranging from Y Combinator companies to Fortune 100 firms. The RL training use case is particularly demanding. Dylan Patel's recent analysis of post-training improvements credits RL environment concurrency as a key factor, requiring orchestration that is not trivial and not something available off the shelf.

The Series A reflects broader market recognition that agents need isolated compute. OpenClaw, which prompted some founders to buy Mac minis for local agent execution, underscored the security and operational constraints of virtual-only architectures. Daytona's pitch is that this demand will scale exponentially as agentic workflows proliferate across enterprises.

Burazin is also organizing Compute, a developer conference at the Chase Center next week with approximately 1,500 attendees, where Patel will speak on RL environments and model improvements.