Your Coding Agents in YOLO Mode… Safely
summary
Guillaume Lours (Docker) explains the risk of “YOLO” agent modes that bypass permission prompts and get full access to a developer’s machine, causing data loss and environment tampering. He describes Docker Sandboxes, which combine VMs and containers so coding agents run in an isolated Docker engine without host access, with outbound network controlled by corporate rules and no access to API keys or credentials, enabling high agent autonomy while protecting the host.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 3:58Agent autonomy risks
- 7:29Docker Sandbox overview
- 10:16Demo: isolated sandbox
- 12:00Demo: SBX Exec security
- 14:29Demo: network rules & audit
- 18:22Demo: secret injection
- 21:56Secret security details
- 24:29Demo: git worktree
- 26:41Current limits & future
- 28:25Q&A
related talks
52mYou Gotta Keep the Dogs Away: YOLO Developer Workflows with a Coding Agent in a Box
34mYOLO Developer Workflows with a Coding Agent in a Box
34mSecure Your Coding Agent Like It’s Malware
27mFantastic Vibes and Where To Find Them
26mrm -rf: Horror Stories From Unsandboxed AI Agents (and How Docker Fixes This)
25mPlease Sandbox your AI Agents
34mZero Secrets: Building Secure Infrastructure for AI Agents
14mTaming Agentic AI: Run Rogue Code Safely on Kubernetes with Agent Sandbox