AIBRIEF
The editor's AI coding agents now run inside a dedicated multi agent window, and the same model that powers them decides which actions need a human's OK.
An emerging trust-boundary pattern in AI agent tools: the same model that runs an agent also decides which of its own tool calls need a human's approval before they execute. VS Code 1.130, released July 22, is a concrete instance of this design. The update wraps the editor's AI coding agents in a dedicated Agents window and gates tool execution through the same model that powers them, letting it self-triage which actions require user approval and which can run. The change is enabled by the setting chat.assistedPermissions.enabled, per InfoWorld's writeup of the release. When it's on, the assistant's underlying LLM evaluates each tool call's risk and either runs it or routes it to a user approval prompt. It's a real step from chat-with-AI toward agents-that-act-with-a-risk-gate, though the model is still triaging risk, not acting without oversight.
The Agents window is the container. It lets developers run and track multiple agent sessions in parallel across projects without opening each workspace in a separate window, and the 1.130 update adds file-level diff statistics and a more compact multi-file diff view so it's easier to scan what each parallel agent changed. Quick chats collapse into single-line rows; regular sessions keep a second line with change counts, status, and timestamps.
Two smaller polish items round out the release: file links now open from Git diff output in the terminal when diff.mnemonicPrefix is on, and chat timestamps appear on hover behind the chat.verbose setting. The release is one week after VS Code 1.129 and otherwise reads as a routine point update.