Microsoft adds AI agent sharing to VS Code

Visual Studio Code is turning into a much more agent-heavy editor. That is the point of the new 1.119 release — not just better autocomplete, but agents that can inspect live browser tabs, ask for access, and keep working with less token burn. But the timing matters. Microsoft is shipping those upgrades right after it had to unwind a Copilot attribution change that made developers feel like AI was claiming credit for human work. (code.visualstudio.com) ### What changed in VS Code this week? VS Code 1.119, released on May 6, centers on “smoother agent interactions” and “more efficient trust and security controls.” The biggest user-facing addition is browser sharing for agents. You can now attach an open browser tab directly into chat, or let an agent ask for access to (code.visualstudio.com)added a lighter-weight model path for managing agent todo lists, which is basically a token-saving trick for long sessions. (code.visualstudio.com) ### Why does browser sharing matter? Because it closes a really annoying loop in web development. Before this, an agent could edit code, but checking whether the fix actually worked in the browser was clumsier. Now the agent can work against a live page, reload it, inspect what changed, and iterate faster. Microsoft is (code.visualstudio.com)ed automatically, and the user has to explicitly approve access. (code.visualstudio.com) ### What is the token optimization actually doing? It is not some huge model breakthrough. It is a workflow optimization. VS Code says it now uses a lightweight model to manage agent todo lists, which reduces the amount of expensive context budget spent on bookkeeping. That fits a broader direction Microsoft started out(code.visualstudio.com)ontext, streaming bulky outputs to temp files, and preserving only the important parts of a session. Basically, the editor is being tuned for long-running agent work, not just one-shot prompts. (code.visualstudio.com) ### So why are people talking about trust? Because Microsoft had just stepped on a land mine. In a GitHub issue posted this week, the VS Code team explained that a setting called `git.addAICoAuthor` originally launched with the default set to off, then changed to `all` in version 1.117, whose public rollout started Apri(code.visualstudio.com)attributed to Copilot, even when AI features were disabled. That meant commit messages could include a “Co-authored-by: Copilot” trailer when Copilot had not actually helped. (github.com) ### How did Microsoft respond? First, it narrowed the default in 1.118, whose public rollout started April 29. Then it fully reverted the default back to `off` for 1.119, with the public rollout starting May 6. Microsoft also said the feature will now stay disabled when `disableAIFeatures` is true, regardless of the at(github.com)onsent before a commit trailer is added. (github.com) ### Why did that small setting cause such a big reaction? Because commit metadata is not cosmetic. It affects provenance, compliance, internal policy, and simple professional credit. A bad autocomplete suggestion is annoying. A tool silently editing authorship metadata is different — that touches ownership. Microsoft e(github.com)ed a possible “assisted-by” approach instead, which would better match what AI tools usually do. (github.com) ### What does this say about Microsoft’s bigger plan? Microsoft clearly wants VS Code to become a control center for agents — across repos, sessions, browser state, memory, and observability. The 1.119 release also adds OpenTelemetry tracing and more trust-related controls, which tells you the company knows these agent(github.com) aggressive. The social license is fragile. (code.visualstudio.com) ### Bottom line? The new browser-sharing features make VS Code agents more practical. But the Copilot attribution rollback is the real lesson. Developers will accept more AI in the editor — turns out they just want explicit consent, accurate credit, and clear boundaries first. (code.visualstudio.com)