VS Code 1.147
Microsoft released Visual Studio Code 1.147 on November 18, 2026, a stable update that pushes the editor further into agent orchestration. The release adds a preview integration that lets agents drive a real browser, an experimental scheduling system for agent requests, persistent drafts, and model routing controls inside the chat experience.
The update follows 1.146, which shipped a week earlier to stable, and continues a pattern Microsoft has held through the second half of 2026: ship agent workflow features in weekly stable builds rather than waiting for a quarterly milestone.
What Changed
The most consequential addition is Browser Use, marked preview. It allows an agent to interact with the user's actual Chrome or Edge browser rather than an embedded view, which extends what an agent can verify inside a running web application. Microsoft's release notes pair the feature with a dedicated setting that governs chat browser access, so teams can disable it centrally.
Scheduling is the second headline. A Send Later option, labeled experimental, lets developers queue local agent requests and follow-ups for later in the Agents window. The accompanying drafts feature means a request no longer has to be sent the moment it is typed. The release highlights note the intent plainly: keep a task in mind without committing it to the agent immediately.
Model selection also changed shape. Version 1.147 introduces model routers and a model comparison view under chat, giving developers a way to route different request types to different models instead of picking one default for everything.
Remote development received updates in the same release, and enterprise controls expanded, including policy settings intended to let organizations constrain agent behavior in managed environments.
Why It Matters
Each feature is small in isolation. Together they describe an editor that is being rebuilt around long-running, supervised agent sessions rather than single completions. Scheduling, drafts, and browser access all assume the same working model: a task exists over time, moves between contexts, and needs to be inspectable while it runs.
That shift has practical consequences for teams. An agent that drives a real browser can reach logged-in sessions and internal tools, which is exactly why the setting exists and why security review now has to happen before rollout rather than after. Enterprises that treat editor updates as routine patching will find policy work attached to this one.
The model routing features matter for cost as much as quality. Routing simple requests to cheaper models while reserving frontier models for complex edits is the standard way to control agent spending, and until now it required external tooling.
What to Watch
Both Browser Use and Send Later ship as preview or experimental, which in VS Code terms means the surface can change between weekly releases. Teams planning to standardize on either should pin their rollout to a tested build and expect churn.
There is also the compatibility question. VS Code 1.147 continues the weekly cadence, so extensions that lag behind on API adoption can break without warning. The release notes archive lists 1.148 in Insiders already, which means the next stable build is roughly a week away.
What's Next
With agent scheduling, browser control, and model routing all landing in the same cycle, the next obvious questions are governance and observability: audit trails for scheduled agent runs, and limits on what browser sessions an agent can reach. Neither appears in the 1.147 highlights. Given the weekly release rhythm, expect incremental answers rather than a single announcement.
Version 1.147 is available now for Windows, macOS, and Linux through the standard update channel.