VS Code 1.116 Ships the Agent Companion App

Microsoft just turned VS Code from a code editor into an agent orchestration platform, and most developers haven't noticed yet.

VS Code 1.116, released April 15, ships three changes that matter. First, Copilot Chat is now built into VS Code natively. No extension install. It's just there. Second, and more important: the Agent Companion App. This is a separate application that runs parallel agent sessions, each isolated in its own git worktree, with inline diff review and direct PR creation. You can kick off multiple agent tasks across different repos and monitor all of them from a single interface.

Third, and this is the one I'm most excited about: agent debug logging. VS Code now exposes full prompt traces, reasoning chains, and decision paths in a dedicated output channel. If your Copilot agent makes a bad edit, you can trace exactly what context it received and where its reasoning went wrong. This is the first time a mainstream IDE has treated agent decision-making as something developers should inspect rather than trust blindly.

The Companion App establishes an architectural pattern I keep seeing everywhere: the IDE becoming a subordinate execution environment while a separate orchestration layer manages the actual work. Claude Code's agent teams do this. Cursor's background agents do this. Now VS Code does it officially. The editor is where code gets written, but the companion app is where work gets planned, delegated, and reviewed.

For my workflow, the debug logging is immediately useful. I've wasted hours trying to figure out why an agent suggestion was wrong. Now I can see the full context window, the system prompt, the reasoning trace. That's a debugging superpower for anyone building with AI coding tools daily.

Update VS Code today. Try the Companion App. Turn on agent debug logging and look at what your tools are actually thinking.