Sandboxes unleash agents by giving them secure, fully functional computers where they can tackle diverse tasks with minimal setup. This talk explores the architectural challenges of building an agent sandbox cloud. We compare runtime isolation technologies and their trade-offs, examine persistence and storage as the next major unlock for agent capabilities, and discuss the key decisions involved in orchestrating and scaling sandboxes.
Abhishek Bhardwaj works on Agent and Reinforcement Learning Infrastructure at OpenAI. He builds systems that enable large-scale model training in RL environments, as well as secure and scalable cloud sandboxes for OpenAI’s agents. Before joining OpenAI, he created Arrakis, an open-source sandbox for AI agents. Previously, he worked at Google on ChromeOS and foundational microVM technologies, and at Replit on core infrastructure and early versions of Replit Agent.
Timestamps
0:00 Introduction and motivation for AI agent sandboxes
1:31 Why AI models need tools and execution environments
3:51 Product-side challenges: Security and the need for sandboxing
6:44 Comparing research vs. product sandbox requirements
8:24 Overview of the three pillars: Runtime, Persistence, and Orchestration
9:05 First principles of Linux execution: System calls and security vectors
11:15 Evaluating fork() and exec models
12:06 Understanding containers: Namespaces and cgroups
16:26 GVisor as an application kernel alternative
18:29 Hardware-level virtualization (Virtual Machines)
20:34 How VMMs (Virtual Machine Monitors) work with KVM
23:16 Evolution of modern VMMs and Rust-based safety
24:32 What defines a "microVM"?
25:43 Orchestrating microVMs via APIs
27:16 Trade-offs of microVMs (performance vs. security)
30:05 The need for persistent storage in agent sandboxes
31:40 Use cases for persistence: Reliability, long-running tasks, and research
34:36 Design choices for disk snapshotting
36:03 First principles of Linux block storage and file systems
37:25 Implementing always-on vs. explicit persistence
41:20 Scaling and orchestrating sandboxes at fleet level
