Skip to content
NewsProduct launch

GitHub Copilot Adds Org-Level Custom Agents and Reasoning Effort Controls to Visual Studio

· by Pondero Newsdesk

The short version

GitHub's August 28 Visual Studio update lets organization owners publish custom Copilot agents across all repos and gives developers a Low/Medium/High effort dial to control how deeply models reason before responding.

GitHub Copilot Adds Org-Level Custom Agents and Reasoning Effort Controls to Visual Studio

GitHub's August 28 Visual Studio update lets organization owners publish custom Copilot agents across every repo under their account, and gives individual developers a per-request dial to control how deeply a model reasons before it answers.

What

Per GitHub's changelog, the August 2026 update ships five new capabilities. The most consequential for teams is organization-level custom agents: GitHub organization and enterprise owners can now publish custom agents that surface automatically in the Visual Studio agent picker alongside each agent's description and org name. Before this update, custom agents in Visual Studio were user-scoped. Teams had no native way to standardize or distribute them across a repository portfolio.

The second addition is explicit reasoning effort controls. Supported models now expose Low, Medium, and High settings in the chat interface. GitHub frames Low as suited for routine questions and High for complex debugging, architecture decisions, or algorithm work. The control maps directly to token usage, so selecting High accepts a longer, more expensive response in exchange for deeper reasoning.

Three other additions complete the release. The model management panel now shows context window sizes, per-token cost details, and capability flags for each available model, with the option to pin favorites or collapse models used less often. The Git agent gained a pre-PR review mode: ask it to review uncommitted changes or specific commits and it surfaces findings inline in the editor, with a navigable list in the Git Changes panel. Both GitHub and Azure DevOps repositories are supported. The update also added a usage view accessible from the Copilot prompt box showing plan details and limit proximity.

All five features are available on every GitHub Copilot plan, from the free tier through Business and Enterprise, per the changelog.

Why it matters

Org-level agent publishing shifts custom Copilot tooling from a per-developer configuration task to a platform capability. A security team can ship a custom agent tuned for vulnerability triage and have it appear in every engineer's VS agent picker without each engineer configuring it. That centralizes maintenance and guarantees consistent prompting across the organization.

The reasoning effort control gives teams a concrete cost lever on Business and Enterprise plans, where per-token billing makes high-effort responses on routine queries an unnecessary expense. Low effort is now a first-class selection rather than a workaround, so developers can reserve deeper reasoning for the tasks that warrant it.

The Git agent review feature competes with standalone pre-commit tools, but its advantage is placement: findings appear in the same view developers use to stage changes, removing a context switch.

What to watch next

GitHub's changelog notes the Insiders channel carries the latest builds ahead of stable releases. The August 26 global model policy (which lets enterprise admins restrict available models org-wide) pairs directly with org-level agents; watch for a unified admin surface that controls both model access and agent distribution in a single policy layer. Also worth tracking: whether org-level agent publishing extends to the Copilot CLI and the Copilot app.

Sources