Giving Agents Computers — Ivan Burazin, Daytona

2026-05-20 · Source: Latent.Space - Www.latent.space · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Robotics & Autonomous Systems · Depth: Advanced, extended

Summary

Daytona, led by CEO Ivan Burazin, has pivoted to providing composable computers, or sandboxes, for AI agents, experiencing 74% month-on-month growth. Originating from CodeAnywhere, a browser-based IDE, Daytona's current offering provides bare-metal, stateful, and long-running compute environments with a 60-millisecond spin-up time for single sandboxes and 75 seconds for 50,000 concurrent instances. Key differentiators include dynamic resizing, Docker-in-Docker capabilities, and preloaded snapshots on NVMe drives for speed. The company is expanding into Windows and macOS sandboxes to address knowledge work locked in legacy applications, targeting a \$10 trillion market opportunity. Daytona currently supports approximately 850,000 daily sandbox runs for its largest customer and maintains a 15% mean utilization due to spiky, unpredictable agent workloads.

Key takeaway

For Directors of AI/ML evaluating infrastructure for agent deployments, Daytona's bare-metal, stateful sandboxes offer critical speed and dynamic scalability for both long-running background agents and spiky RL/eval workloads. Your teams can achieve significantly faster spin-up times and greater operational flexibility, especially when integrating with existing legacy applications via Windows/Mac sandboxes, which is crucial for unlocking new automation value.

Key insights

AI agents require composable, fast, stateful, and dynamically resizable compute environments distinct from human-centric infrastructure.

Principles

• Agent compute demands bare-metal speed and statefulness.
• Unpredictable, spiky agent workloads necessitate flexible capacity planning.
• Open source can enhance context for agent integrations.

Method

Daytona utilizes bare-metal machines, a custom scheduler, and NVMe-based preloaded snapshots to deliver high-speed, stateful, and dynamically resizable sandboxes with Docker-in-Docker support.

In practice

• Deploy AI agents in sandboxes for rapid, stateful execution.
• Leverage Docker-in-Docker for complex agent workloads (e.g., K3S).
• Consider Windows/Mac sandboxes for automating legacy knowledge work.

Topics

• AI Agents
• Compute Sandboxes
• Bare-Metal Infrastructure
• Dev Environments
• RPA
• Workload Management

Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, MLOps Engineer, Director of AI/ML

Related on AIssential

• Daytona raises $24M Series A to build agent-native compute infrastructure — Agent-native compute infrastructure is crucial for scaling stateful, long-running AI agent workloads beyond traditional cloud models.
• Every Agent Needs a Box — Aaron Levie, Box — Secure, sandboxed environments are crucial for AI agents to operate effectively and safely within enterprise data systems.
• OpenClaw: A Technical Guide for the Business Executive — OpenClaw integrates frontier AI models with system primitives to create persistent, proactive agentic computing.
• Build long-running MCP servers on Amazon Bedrock AgentCore with Strands Agents integration — Persistent state management enables AI agents to execute long-running tasks reliably across disconnected sessions.
• Scaling Agents on Kubernetes with acpx and ACP — Onur Solmaz, OpenClaw — ACP standardizes agent-client interaction, enabling efficient, reusable agent integrations and automated workflows.
• The Minimum Viable AI Agent Server — AI agent orchestration performance is more dependent on the agent's approach and software stack than on dedicated local hardware.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Latent.Space - Www.latent.space.