Visual Studio Code Agents

Visual Studio Code Agents is a coding tool. Agent features inside VS Code that plan, edit, run commands and iterate on whole coding tasks. Work runs in parallel across an Agents window, the editor chat and a Copilot CLI terminal, with approvals and model choice left to the developer.

Visual Studio Code Agents is the agentic development layer built into VS Code. Instead of suggesting single lines, an agent takes a goal, inspects the repository, drafts a plan, edits files, runs commands and uses build or test failures to decide what to do next. The developer chooses how much direction and review each task gets, and can approve sensitive actions, inspect changes as they land and redirect the work without losing progress. The feature is organized around three working surfaces. An Agents window tracks local, background and cloud sessions so several jobs, such as a bug fix, a refactor and a dependency upgrade, can run side by side with separate context and progress. The editor offers a Chat view for a code-first workflow, where edits can be watched live, follow-up questions asked, and control taken back at any moment with IntelliSense, refactoring, source control, testing and debugging still available. The terminal route uses the Copilot CLI, installed through npm, so agent work can start in the shell and later be opened in VS Code for closer inspection. Typical uses listed include building a feature end to end, tracing and fixing failing tests, planning framework or API migrations, turning a Figma file or rough idea into a reviewable pull request, auditing accessibility by running the app, and drafting documentation or release notes. Behavior can be shaped with instructions, skills, custom agents, hooks and plugins, and the agent harness, model, provider or local endpoint can be selected. Longer jobs can be handed to the cloud and returned as pull requests. It suits professional developers and teams who already work in VS Code and want agent autonomy without leaving their editor. Compared with standalone AI editors or terminal-only coding agents, the main distinction is that planning, parallel sessions, the editor and the command line all sit in one familiar environment, with model choice left open rather than tied to a single provider.