GitHub Copilot in Visual Studio Gets Thinking Controls and Org Agents

GitHub shipped its August 2026 update for Copilot in Visual Studio, adding five notable features: organization-level custom agents, adjustable model thinking effort (Low, Medium, High), better model management with pinning and a detailed capability view, a usage dashboard accessible from the prompt box, and Git agent reviews of uncommitted changes and commits. The update is available across every Copilot plan, from Free to Enterprise.

What happened

Feature	Detail
Org-level custom agents	Published by org/enterprise owners, auto-detected in Visual Studio
Thinking effort settings	Low, Medium, or High per supported model
Model management view	Shows capabilities, context window sizes, and cost info
Usage dashboard	Accessible via “View all Copilot usage” in the prompt box
Git agent review	Reviews uncommitted changes and commits inline, works with GitHub and Azure DevOps
Plans covered	Free, Student, Pro, Pro+, Max, Business, Enterprise

Organization-level custom agents

GitHub organization and enterprise owners can now publish custom agents that are available across every repository in their org. Visual Studio picks these up automatically, showing each agent’s description and its organization source in the agent picker. You do need an active GitHub organization for this to work.

Adjustable thinking effort

Supported models now expose three thinking effort levels. Low is suited for routine tasks like boilerplate or simple edits. Medium covers everyday coding. High is intended for complex debugging, algorithm design, and architecture decisions where deeper reasoning is worth the extra token cost. This lets teams control spend without switching models entirely.

Model management and usage visibility

The updated model picker lets you pin favorites, collapse less-used options, and open a dedicated management view. That view lists each model’s capabilities, context window size, and cost details alongside controls for both Copilot and custom models. From the Copilot prompt box you can also select “View all Copilot usage” to see your full plan status, and refined notifications now tell you clearly when you are close to a plan limit and what your options are.

Git agent code review

The Git agent can now review uncommitted changes or whole commits before you open a pull request. Findings show up inline in the editor and in a navigable list inside the Git Changes panel. You can continue the chat conversation to dig into any suggestion or address it directly. According to GitHub, this works with both GitHub and Azure DevOps repositories.

Why it matters

The thinking effort control is the most practically useful addition here. Token costs are real, and a Low setting on a fast model can handle most day-to-day edits far cheaper than running everything at maximum reasoning depth. Teams paying for Business or Enterprise plans will want to set a sensible default to avoid budget creep.

Org-level custom agents close a gap that forced teams to share agent configurations manually or through workarounds. Now a single publish by an org owner propagates to every developer without extra setup, which matters for teams that have built specialized agents for internal APIs or coding standards. If you are thinking about building those kinds of agents, our AI integration work covers exactly this territory.

The Git agent review feature is a quieter win. Catching issues before a pull request is opened, rather than after reviewers have been assigned, fits naturally into a solo or small-team workflow where formal PR review cycles are a bottleneck.

Our take

Thinking effort controls are a genuinely good idea, but the framing of “Low for simple tasks, High for complex ones” places the judgment on the developer. Most people will leave it on High and not think about it, which defeats the purpose. GitHub should consider making token cost feedback visible per-prompt so developers actually see the difference. The model management view with cost information is a step in that direction, and we hope GitHub makes that data more prominent over time.

The org-level agent publishing is the feature we are most likely to use with clients. A shared internal agent for, say, a company’s design system or API layer is far more valuable than a generic assistant. We have covered similar themes around GitHub Copilot Enterprise’s plugin marketplace auto-updates, and that ecosystem is clearly maturing.

If you are running a small dev team and not yet on a Copilot plan, the Free tier now includes enough to evaluate most of these features before committing.

What to do about it

1. Update Visual Studio to the latest release (check the Insiders channel for the newest build).
2. Open the Copilot prompt box, select “View all Copilot usage,” and note your current plan limits.
3. Switch your default model thinking effort to Low or Medium and reserve High for specific debugging sessions.
4. If you manage a GitHub organization, review whether any internal agents are worth publishing org-wide through the new custom agent workflow.
5. Try the Git agent review on your next set of uncommitted changes before opening a pull request.

The clearest near-term win is pinning a cost-efficient model at Medium thinking effort as your default and using the usage dashboard to verify that token spend stays in line with expectations.

Source: GitHub Changelog