FOR THE BUILDERS, THE MAKERS & THE SMALL TEAMS WITH BIG IDEAS ✳
AI Coding · News Analysis

GitHub Copilot’s Experimental Inbox and Model Comparisons: What to Evaluate

GitHub’s November 16, 2026 Copilot roundup highlights two experimental additions in VS Code 1.146: an Inbox that surfaces agent sessions needing attention across workspaces, and tools that let agents run model comparisons. For solo developers juggling coding tasks, these features deserve separate evaluations. Try Inbox if finding the next session to review is becoming difficult; try automated comparisons when you have a specific model-selection question and clear acceptance criteria.

ai-coding: Official update → Team impact → Adoption decision. Editorial workflow illustration.
Source-based News Analysis. GitHub’s roundup was published November 16, 2026; VS Code 1.146 release notes are dated November 11, 2026. Both official sources were accessed November 17, 2026. We did not run VS Code, exercise either feature, or measure performance. Feature descriptions are vendor-documented; evaluation steps and adoption recommendations are editorial proposals.

1. Separate the roundup date from the release

The GitHub Copilot weekly roundup was published November 16, despite its “November 9” title. It describes features in VS Code 1.146, whose release notes give a November 11 release date. The roundup brings these changes together; it does not establish that they first shipped November 16.

There is also a useful distinction in maturity labels. VS Code describes the broader model-comparison feature as Preview, while the tools that let agents run comparisons are Experimental. Inbox mode is Experimental too. Neither experimental addition should become essential to a delivery workflow before you evaluate its behavior.

This article analyzes those official descriptions. We have not operated the features or verified their reliability through hands-on testing. The practical checks below are editorial proposals, not reported results.

2. Inbox organizes attention across workspaces

According to the release notes, Inbox is off by default. Enable sessions.list.inboxEnabled, then select Grouping > Inbox Mode in the Agents window’s sessions list. Sessions needing attention appear above the Working, Ready to merge, On hold, and Done groups.

GitHub says this view tracks sessions across all your workspaces. That makes it relevant when a developer alternates between several projects and must decide which agent needs a response next. The potential benefit is easier triage; the sources provide no independent measurement of time saved or missed interruptions avoided.

The available actions include Dismiss from Inbox, Put on Hold, and Snooze, with Undo available if you change your mind. Dismissal does not archive a session. The notes direct users to Mark Session as Done for archiving. That distinction matters when clearing the list: setting work aside and finishing it are different decisions.

3. Agent-run comparisons add another layer of delegation

VS Code’s Compare Models feature lets you run the same prompt across models, evaluate responses with a judge, and combine their strengths with a synthesizer. Version 1.146 adds experimental tools that allow agents to discover models, launch comparisons, and read results and conversation references to summarize findings across scenarios.

The tools require Compare Models to be enabled and use sessions.chat.compareAgents.tools.enabled, which is off by default. Disabling those tools leaves manual comparisons available. A developer can therefore assess the comparison workflow manually before delegating its operation.

The release also allows a judge and synthesizer for chat-only responses without selecting a workspace. Selecting Synthesize creates a new chat without changing the original attempts. These are documented capabilities, not evidence that a synthesized answer is more accurate. Keep the original attempts available when examining whether the summary preserved important disagreements.

4. Give the comparison an answerable question

Our proposed first comparison would use one bounded task, such as explaining a failing unit test and proposing a minimal correction. Give each attempt the same starting material, expected behavior, and constraints. Record the models and any effort settings used so the result remains interpretable later.

Define the decision before requesting a judge’s assessment: Does the explanation identify the cause? Does the proposed patch address the failure? Does it introduce unrelated changes? A persuasive explanation alone cannot establish correctness. Our guide to regression tests for AI-generated code explains how to connect generated changes to checks you can actually run.

Ask the summarizing agent to identify the evidence behind its conclusions and preserve unresolved differences. An automated judge’s preference remains an assessment to inspect. A developer should review the actual changes and relevant test output before accepting a recommendation.

Keep the first comparison small and stop once it answers the stated question. Additional attempts are useful when they resolve uncertainty; a larger collection of responses can also create more review work. This article offers no measured model ranking.

5. Evaluate Inbox without increasing your workload

For an Inbox pilot, use the sessions you already manage across a few workspaces. Before enabling it, note how you currently find stalled conversations and completed work. Then check whether Inbox makes those same decisions easier during an ordinary work period. Avoid launching extra tasks merely to populate the interface.

Try dismissing or postponing a nonurgent session, and verify that you can find it again when needed. Record any confusion between setting work aside and marking it done. These are proposed usability checks; we have not verified snooze timing, state persistence, or cross-workspace completeness.

Keep code acceptance separate from list organization. A session appearing under Ready to merge still deserves review of its diff and checks. Use our AI code review workflow to define that acceptance step independently of how sessions are displayed.

6. Adopt each feature against a concrete problem

Inbox is worth a limited trial when navigating concurrent sessions already interrupts your work. If you normally run one task at a time, its immediate value may be modest. Judge it by whether you can identify and resume the right session with less confusion.

Agent-run comparisons suit developers who already know what they want to compare and how they will check the outputs. Start manually if those criteria are still unclear. Continue with automation only when its summaries remain traceable to the underlying attempts and help you make a defensible decision.

The release expands how developers organize and evaluate agent work. Whether it improves your workflow depends on the review effort it creates as well as the navigation or comparison work it removes.

Sources & further reading

Primary references used for product and technical information. Features and terms can change; check the provider before committing.

No affiliate links are included in this article. Read our disclosure.