AI Sandboxes: Running Coding Agents Safely in Production-Grade Environments
-
Kristiyan Velkov
Jakala
Principal Frontend Developer
-
Shivay Lamba
Couchbase
Senior Software Engineer
As coding agents like Claude Code, OpenAI Codex and Gemini CLI become part of developers’ daily workflows, one problem looms large: we’re letting them run wild on our machines (or put ourselves into a Skinner box, playing an agent-approval slot machine game). The ideal workflow is “YOLO mode”. where an agent can freely edit, build, and test code. But that freedom shouldn’t come at the cost of safety or control.
In this talk, we’ll explore how Docker is building an opinionated developer experience for running coding agents safely inside sandboxes, combining the isolation benefits of containers with the seamless UX developers expect. We’ll discuss what we learned from Devcontainers (and why they never saw broad adoption), and how sandboxes evolve those ideas to make “safe by default” not just possible, but pleasant.
You’ll leave with a new mental model for local coding agent based development: fast, flexible, and fearless, the YOLO workflow done right.
World Congress 2026 Europe
Kristiyan Velkov
Jakala
Principal Frontend Developer
Shivay Lamba
Couchbase
Senior Software Engineer
Ben Hutchison
Oplane
Forward Deploy Engineer
Carl Fagerlin
Oplane
Forward Deploy Engineer
Anshul Jindal
NVIDIA
Senior Solution Architect
Tsubasa Onishi
NVIDIA
Senior Solutions Architect
Guillaume Moigneu
Upsun
Field CTO