Descriptions:
In this Latent Space podcast episode, host Alessio Fanelli sits down with Ivan Burazin, CEO of Daytona, to discuss the explosive growth of AI agent infrastructure. Daytona provides sandboxed compute environments for autonomous agents, and Burazin shares that the platform is now seeing 74% month-over-month growth and processing 850,000 sandbox runs per day — numbers he attributes to a structural gap in the market for fast, developer-friendly execution environments purpose-built for agents rather than adapted from general container orchestration.
Burazin traces Daytona’s lineage from CodeAnywhere, one of the earliest browser-based IDEs, through a pivot into developer conferences before refocusing on agent infrastructure. He explains how Daytona differentiates from managed Kubernetes offerings like Amazon EKS and Google GKE: a Stripe- and Twilio-style consumption API, sub-second sandbox spin-up times, and dynamic on-the-fly memory resizing that makes OOM errors rare — a capability he describes as nearly impossible on conventional Kubernetes deployments. The platform’s declarative image builder lets agents specify dependencies at runtime, with snapshots propagated across parallel runs.
The conversation also covers how reinforcement learning workloads have grown to roughly 50% of Daytona’s traffic, the TerminalBench team’s unprompted recommendation of Daytona in their framework documentation, and the launch of Agent Cloud — a new product targeting teams that need scalable, isolated agent execution without building on raw container orchestration. Burazin positions the long-term market as encompassing every agent that will ever run, a framing that underpins Daytona’s infrastructure-layer bet.
📺 Source: Latent Space · Published May 21, 2026
🏷️ Format: Interview
