Docker Sandboxes puts each coding agent in its own disposable microVM
Docker's new sbx CLI boots each AI coding agent into a disposable microVM with a dedicated kernel and private Docker daemon, mounting only the project workspace so agent-executed code cannot reach the host. That turns --dangerously-skip-permissions into a bounded risk, and six agent CLIs — Claude Code, Codex, Gemini CLI, and three more — can run unattended on macOS or Windows today. Docker wrote its own VMM rather than reusing Firecracker so it runs natively on all three desktop hypervisors; the core product is free, with team-wide policy enforcement in the paid Docker AI Governance add-on.
Source: docker.com ↗
YOLO mode (--dangerously-skip-permissions) gives agents autonomy with no approval prompts. Essential for speed, but risky without guardrails. Sandboxes make it safe by isolating each agent inside a dedicated microVM.
Why this matters
- → Lets AI coding agents run untethered without host filesystem risk
- → Turns disposable microVMs into the standard execution boundary for six agent CLIs
- → Teams can enforce sandbox policies uniformly across developers