GitHub made Copilot's browser tools in VS Code generally available on July 1 and turned them on by default. More than a convenience feature, it is an architectural choice. The feedback loop that previously ran from writing code, to manually testing it, to pasting the error back into a chat window now closes itself. The agent covers the middle step.
What agents can do now
In practice a Copilot agent can open a page after generating code, walk through a user flow the way a real user would, read console output, capture screenshots, and write a fix based on what it sees — all without you acting as the human relay in between. Navigation, clicking, typing, hovering: the capability set is serious enough to catch real UI bugs, not just trivial rendering issues.
Privacy as a first-class decision
The privacy design deserves attention. Your own browser tabs stay fully shielded: the agent cannot see or interact with them unless you explicitly choose Share with Agent. Tabs the agent opens itself run in isolated sessions with no access to your cookies or local storage. These are not afterthoughts — they are the defaults from day one. A tool that operates this close to your live browser warrants exactly this kind of clear boundary.
Our take
This is the most concrete step forward for AI in the development workflow this year. For a long time the story was faster typing and smarter autocomplete; now the agent helps verify that code actually works in the browser. That is a qualitative shift, not a gradual one. At the same time: the agent verifying things is not a reason to review your own work less carefully. Agentic coding means shorter feedback loops, not less responsibility for what you ship.
