Coding agents like Claude Code, Gemini CLI, and others are changing the development landscape. These tools have gained autonomy, allowing them to perform actions like deleting repositories and accessing sensitive information. However, this increased capability raises security concerns. Developers need a way to provide these agents with enough access to be useful while safeguarding their local environments from potential risks. Docker is addressing this issue by introducing a sandboxing approach. Their solution involves running coding agents in isolated local environments using containers. This setup mimics the developer’s workspace while enforcing strict boundaries, which helps reduce the chance of unintended consequences on the local system. Currently, this experimental feature operates within Docker Desktop's virtual machine but will eventually transition to dedicated microVMs for enhanced security. The experimental preview includes features like container-based isolation, filesystem protection, and support for popular coding agents. To use it, developers need Docker Desktop version 4.50 or later and can initiate a new isolated environment with a simple command. Despite being in an early stage, Docker plans to improve this offering by adding support for multiple agents, better network controls, and centralized management. They are actively seeking feedback from users to refine the experience.