Daytona CEO谈AI Agent需要专属沙盒Daytona CEO on Why AI Agents Need Dedicated Sandboxes
核心要点：每个高效的AI Agent都需要自己的专属计算机（沙盒），才能作为数字知识工作者发挥作用。Daytona CEO兼开发者工具老兵Ivan Burazin认为，为Agent提供隔离、有状态的环境是实现超越简单对话的实际效用的基础。Agent需要沙盒来保障安全、并发、长时间运行任务，以及访问缺乏headless API的遗留系统。超大规模云为无状态应用而建，不够用——沙盒需要新的原语，如快速启动（60毫秒）、实时调整大小、快照，以及支持不同OS/GPU。Burazin解释，从临时到长时间有状态沙盒的转变需要从头重构调度器。他还预测随着Agent基础设施扩展，CPU短缺即将到来。一句难忘的话：“我的观点是每个Agent至少需要一个沙盒，有时更多。” Daytona与主要Agent框架合作，强调基础设施、原语和工具构成完整沙盒产品。
The Takeaway: Every productive AI agent needs its own dedicated computer (sandbox) to function as a digital knowledge worker. Ivan Burazin, CEO of Daytona and veteran builder of developer tools, argues that giving agents isolated, stateful environments is fundamental for real-world utility beyond simple chat. Agents require sandboxes for security, concurrency, long-running tasks, and accessing legacy systems that lack headless APIs. Hyperscalers built for stateless apps don't suffice—sandboxes need new primitives like fast spin-up (60ms), live resizing, snapshots, and support for diverse OS/GPU needs. Burazin explains the shift from ephemeral to long-running stateful sandboxes required rebuilding their scheduler from scratch. He also predicts upcoming CPU shortages as agent infrastructure scales. A memorable quote: "My argument is that every agent will need at least one sandbox, sometimes more." Daytona partners with major agent frameworks and emphasizes infrastructure, primitives, and tooling as the full sandbox offering.
查看原文 →