GitHub shipped a wave of Copilot updates at Microsoft Ignite on 17 November 2026. The most notable is GitHub Copilot Evals, now in public preview, which lets developers write repeatable tests for their own Copilot skills, custom agents and MCP tools. Copilot code review also gains an Adaptive effort mode and secret detection, and browser use arrives across Copilot’s tools.
Quick Take
- Copilot Evals: public preview for Copilot Individual, Business and Enterprise; run with the /eval command or in plain language.
- Adaptive code review: Copilot picks Lite, Balanced or Max review depth per pull request; public preview.
- Secret detection: code review flags exposed credentials, using 500+ known secret types plus AI detection; public preview.
- Browser use: Copilot can control a Chromium browser via the Copilot Chrome extension, in CLI, the Copilot app, VS Code and JetBrains; public preview.
- Azure DevOps: Copilot code review for Azure Repos is generally available; the Copilot app gains Azure DevOps repositories in preview.
In plain English
Developers increasingly customize Copilot with their own instructions, agents and tools, but it is hard to know whether a change made things better or worse because AI answers vary from run to run. Copilot Evals is a test kit for exactly that: you describe a realistic task and what a good result looks like, run it, and compare before and after. Separately, Copilot’s automatic code reviews now decide for themselves how deeply to check each change and warn when someone accidentally commits a password or key.
What happened
GitHub published more than a dozen changelog entries tied to Ignite. In the Evals post, GitHub defines an eval as “a repeatable test for AI behavior”: “You give an agent a realistic task, describe what a good result looks like, and check both the result and how the agent got there.” Each run must be approved by the user before it starts. Evals uses Vally, an extensible local eval tool, as its engine. The same day, GitHub added MAI Code 2 and MAI Code 2 Flash to Copilot.
How it works
- Evals: create an eval for a skill, custom agent or MCP tool, run it, see which checks passed or failed, find out why, and compare runs. Install via the VS Code extension, the Copilot CLI plugin or the github/evals marketplace in the Copilot app.
- Adaptive: Copilot reads each pull request’s complexity and risk and shows which effort level it chose. It never goes above the maximum set by the enterprise, organization or repository. Balanced stays the default.
- Cross-repository context: code review can read related public repositories to find code that depends on a change; public preview for public repositories.
- Secret detection: combines deterministic detectors from GitHub’s secret scanning partner program with LLM-based detection for secrets without a recognizable format.
- Browser use: off by default; runs in an isolated tab group and can fill forms, pull information from pages and test web apps.
- Event Streams: a new preview lets agents keep a connection open to receive pull request and check updates instead of polling, behind a waitlist.
Why it matters
For developers, Evals turns “it seemed better” into something measurable, which matters as teams pile up custom agents and MCP tools.
For engineering leaders, Adaptive review and secret detection move Copilot from a helper into a gate in the merge process, with limits that admins still control.
For security teams, catching leaked keys at review time addresses a real problem: coding agents have already been tied to secrets leaking into public repositories.
What’s still open
- How much an eval run costs in AI credits for larger suites.
- How often Adaptive picks Lite for a change that deserved Max.
- Whether browser use, with access to signed-in sessions, brings new phishing and permission risks.
Sources
- GitHub Changelog, “GitHub Copilot Evals is now in public preview,” 17 November 2026
- GitHub Changelog, “Copilot code review: Adaptive effort, secret detection, and more,” 17 November 2026
- GitHub Changelog, “GitHub Copilot app, Microsoft Ignite releases” and “Let GitHub Copilot interact with your browser,” 17 November 2026
- GitHub Changelog, “GitHub Event Streams is now in preview,” 17 November 2026
i10x Perspective
The most useful item here is the least flashy. As every team builds its own agents, a standard way to test them inside the tool developers already use could matter more than another model. Follow more in our Coding section.






