From fork() to Fleet: Designing an Agent Sandbox Cloud Pt2
Summary
Abhishek Bhardwaj's World's Fair session is a Sandbox & Platform Engineering talk about the infrastructure layer that lets agents operate inside secure, real computing environments rather than toy execution sandboxes. The official session description frames the problem at cloud scale: compare runtime isolation technologies, decide how much kernel and process boundary protection is needed, make persistence and storage usable for longer-running agents, and orchestrate many sandbox instances as a production fleet. The connected speaker page adds useful context: Bhardwaj works on RL and agent infrastructure at OpenAI, so the session should be read as coming from the layer where agent training, tool use, and execution environments meet. The linked Arrakis video and slide pages are supporting evidence from a related public AI Engineer talk by the same speaker, not a confirmed recording of this exact World's Fair session, but they make the technical vocabulary concrete: Linux namespaces, containers, process cloning, mounts, filesystem snapshots, userspace behavior, syscalls, kernel boundaries, and attack surfaces. Together, the schedule and Arrakis materials point to a systems design talk about moving from a single sandbox runtime toward a fleet-scale agent sandbox cloud: how to isolate untrusted or semi-trusted agent work, how to preserve state without weakening security, and how to make sandbox orchestration reliable enough for production AI engineering workflows.
Official Schedule Context
- Date/time: 2026-06-30 · 1:55pm-2:15pm
- Track/room: Sandbox & Platform Engineering · Track 1
- Speaker(s): Abhishek Bhardwaj
- Session type/status: session · confirmed
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
- Pending transcript synthesis when an official recording or confirmed matching video is available.
Supporting Slides
- youtube wsFd22SL1s8 slides — extracted from the related public AI Engineer video.
Slide Evidence
- Slide-only cropped deck: youtube wsFd22SL1s8 dense slides (42 viable slide images).
- Related slide/OCR pages:
- youtube wsFd22SL1s8 dense slides
- youtube wsFd22SL1s8 reconstructed slides
- youtube wsFd22SL1s8 slides
- Slide-derived terms:
namespace,arrakis,process,containers,kernel,container,linux,code,chatgpt,clone,mount,attack,snapshot,server,python,userspace,version,syscall