What happened
GitHub rolled up its Copilot changes across VS Code versions 1.123 through 1.127, covering everything shipped through June and early July 2026. Agentic browser tools are now generally available, and parallel sessions include credit tracking across delegated work. The model picker also lets users discover and install providers inside VS Code.
Why it matters
The integrated browser can now be driven by the agent itself, not just by you. Copilot can navigate pages, inspect content, take screenshots, and check that a web app actually works, directly inside VS Code. You can also hand it context by dropping in a screenshot or an area screenshot in chat, useful when a bug is visual and hard to describe in words.
That pattern is no longer specific to Copilot. Anthropic added an in-app browser to Claude Code Desktop on July 11. Claude can open documentation, designs, and external sites alongside a running app. It can inspect the page and interact with it while working on the code.
Copilot and Claude Code now let agents inspect and interact with pages in built-in browsers. An agent can move from writing code to checking the interface without waiting for the builder to describe every visual problem.
Parallel sessions matter more than they sound. Instead of one agent working through one task while you wait, you can run several sessions side by side, each with its own focused chat. One session implements a feature while another reviews a prior change or writes tests, and you check on each without losing track of the others.
Cost visibility finally covers what you’re actually spending. Previously you could see per-request costs. Now the status dashboard tracks credit usage across a full chat session, including work delegated to helper agents. Credits are the units drawn from your account’s usage allowance, so you can track the whole task rather than its final message.
The Language Models editor now supports direct discovery and installation of model provider extensions, plus one picker for context size, how much information a model can work with, and reasoning effort, how much thinking it spends on the task. That’s a meaningful change if you’ve been manually hunting for the right model or provider extension for a given task.
GitHub says Autopilot also improved at recognizing when a task is done, reducing manual prompts to continue.
Who should care
Anyone already running Copilot as an agent inside VS Code, not just using it for inline suggestions. If you’ve been running one agent task at a time and waiting on it, parallel sessions change your workflow directly. If you’ve ever been surprised by your Copilot bill, the new cost dashboard is worth checking before your next billing cycle.
Claude Code Desktop users should pay attention to the same browser shift. The feature is most useful when an agent needs to verify a web flow, inspect documentation, or compare its implementation against a design.
Teams juggling multiple models or providers will get the most out of the Marketplace model discovery and the unified context/reasoning picker, since switching used to mean digging through settings.
What builders should do next
Update VS Code, then open the integrated browser panel and try letting an agent navigate a page on its own, rather than screenshotting it yourself first.
If Claude Code Desktop is your main agent, update the app and repeat the same test there. Ask Claude to open a running page, complete one important flow, and report what it verified.
Check your Copilot spending in the status bar dashboard before you start a new task, so you have a baseline for what a normal session costs. If a task starts running unusually long, that dashboard can show how usage is growing.
Try running two agent sessions side by side on a real task split, like implementation in one and test writing in the other, and see whether that changes how you’d normally sequence the work.
End of article