Docker moves AI agent sandboxes into managed cloud
Docker announced Docker Cloud Sandboxes at WeAreDevelopers North America, allowing AI agent workflows to keep running in the cloud after a developer’s laptop is shut down. The service extends Docker’s local sandbox isolation to Docker-managed cloud infrastructure while preserving the same CLI, trust model, Kits, microVM isolation, and policies. It is intended to let organizations run agentic workloads at scale without tying up developer hardware, provisioning their own infrastructure, or paying for unused capacity.
Docker unveiled Cloud Sandboxes at WeAreDevelopers North America. The offering carries local sandbox protections into Docker-operated cloud infrastructure, so agent tasks can persist after a developer’s machine powers down. It keeps the same command-line interface, trust approach, Kits, microVM separation, and policy controls.
The service targets agentic workloads that need to run unattended for long periods or scale in parallel. Docker manages capacity from 1 to 16 vCPUs, with fast startup and built-in secrets, policy, MCP gateways, and agent configuration. Organizations can avoid dedicating laptops or provisioning their own infrastructure.
Docker Cloud Sandboxes could shift more agent execution from personal devices to managed cloud environments. Developers may gain freedom from keeping laptops awake, while teams could run longer or parallel tasks with less infrastructure work. This may accelerate adoption of autonomous agents in software and operations, but also concentrate more sensitive workloads and policy enforcement in cloud providers, making governance, cost oversight, and access controls increasingly important.