In Brief: Autonomous AI agents require provisioned, stateful compute sandboxes instead of local machines, creating a new market for specialized infrastructure providers like Daytona.
Daytona, an AI infrastructure provider, positions itself as the central platform for autonomous agents that rely on provisioned compute sandboxes. The rationale behind the approach: agents cannot be paused and require reliable, API-driven resources instead of local development machines.
The market for AI agents has triggered an infrastructure transformation. As products like Perplexity, Manus, and Cursor increasingly equip agents with computer access, parallel evaluation infrastructures like TerminalBench and GDPVal are expanding. Daytona benefits from the establishment of a standardized LLM OS stack – and only a few specialized infrastructure providers serve this market.
Daytona is based on a vision more than a decade old, held by CEO Ivan Burazin: the end of localhost development. His earlier company CodeAnywhere (2010) attempted to move development environments into the browser – the market was not ready at that time. Agents fundamentally change the playing field: they disregard laptops and local editors; instead, they require an API-accessible, stateful computer that is immediately available, dynamically scalable, and runs in isolation.
Daytona achieves this using bare-metal hardware with its own scheduler. The company spins up individual sandboxes in approximately 60 milliseconds and can provision 50,000 sandboxes in around 75 seconds. A major customer runs approximately 850,000 sandboxes daily. Notably: reinforcement learning and evaluation workloads have grown from 0 percent to approximately 50 percent of total usage in just a few months – often with extreme load spikes ranging from zero to 100,000 CPUs.
For practitioners, this has concrete implications: agents need not only code execution boxes, but composable compute systems with state management. This also includes Windows and macOS environments. At the same time, it becomes clear that classical orchestration tools like Kubernetes are problematic for these workloads. The emerging market might function less along AWS, EKS, or GKS models, but rather following the example of platforms like Stripe – with clear API boundaries and managed complexity.
Source: ainews-dev.lumi-systems.io · Published May 21, 2026
Lumi AI News — AI-assisted curation in accordance with Art. 50 EU AI Act. Paraphrasing and classification by Lumi News Pipeline v1.5.2.