SummaryStarting with Visual Studio 2026 v18.12, Copilot Chat is transitioning to the Microsoft Copilot SDK framework developed by GitHub. Microsoft claims this change results in shorter, more focused responses and solves up to 21 more out of 157 test tasks. New interactive features include mid-task steering, sub-agent visibility, folder-level context, and built-in agents such as /security-review and /rubber-duck. |
Why it mattersThis move unifies the Copilot experience across VS Code, GitHub CLI, the GitHub app, and now full-featured Visual Studio under a single proven framework, reducing fragmentation for .NET and enterprise developers. The 'same model, better framework' premise is significant because it demonstrates that gains stem from orchestration and tooling rather than raw model capability. This could lower the barrier for teams already using Copilot subscriptions. It also signals Microsoft's strategy to consolidate its AI agent infrastructure under the GitHub SDK rather than building separate stacks per product. |
Highlights- A 157-task benchmark (C#, C++, Python, .NET Framework/Core, Azure, WinForms) showed up to 21 additional tasks solved compared to the old engine
- The framework ensures Copilot works until the task is truly complete, prevents early stopping, and makes per-request costs visible
- New features include mid-task steering for second opinions, sub-agent activity views, folder context, and built-in agents (/security-review, /rubber-duck)
|
Critical takeThe 21-task improvement was self-reported by Microsoft in its own benchmark. Since independent verification or task difficulty distribution was not disclosed, the 'up to 21' headline figure may be selectively presented. The article also conflates marketing language ('proven at scale with millions of users') with engineering evidence. The 'framework' narrative risks obscuring whether the underlying models were also quietly updated between the two test runs. |
This critique was written by the local AI model
Qwen3.8-27B. I chose the story; I do not review each critique before it is published.