From fork() to Fleet: Designing an Agent Sandbox Cloud Pt 1

Summary

Abhishek Bhardwaj's session is about the infrastructure that lets coding and computer-use agents operate inside real, stateful computers without handing them the host machine. The connected Arrakis slide material grounds the talk in a bottom-up sandbox design: starting from process creation with fork()/clone, then layering Linux namespaces, mounts, containers, syscall boundaries, userspace/kernel separation, snapshots, and attack-surface reduction into an environment an agent can safely use. That makes the session less a generic cloud-scaling talk than a concrete tour of how low-level Linux isolation becomes an agent platform primitive.

The platform question is how to turn those primitives into a fleet of useful sandboxes: agents need filesystems, Python processes, servers, networked tools, persistence, and replayable state, while operators need hard boundaries around untrusted code and predictable orchestration at scale. Bhardwaj's OpenAI agent-infrastructure background and the linked Arrakis deck point to the key design tensions: containers versus stronger runtime isolation, snapshots as both product capability and operational control, storage as the next unlock for long-running agent work, and orchestration choices that decide whether sandboxes feel like disposable processes or durable computers for agents.

Official Schedule Context

Official Description

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.

Related YouTube Video

Arrakis: How To Build An AI Sandbox From Scratch - Abhishek Bhardwaj, OpenAI (speaker-match related prior/adjacent AI Engineer video; captions: English auto-captions).

Transcript Status

Related video transcript availability: English auto-captions. Treat this as supporting context, not a recording of this exact scheduled session unless later confirmed. Not fetched yet.

People

Notes

Supporting Slides

Slide Evidence