The Same Brain Behind the CLI Now Comes to Visual Studio — Copilot's New Agent Aligns on the SDK, and Skills Ship "Off by Default"

On July 30, GitHub published its monthly update for GitHub Copilot in Visual Studio. The headline is a new "Agent (Preview)" added to Copilot Chat. It is built on top of the same Copilot SDK that powers GitHub Copilot CLI, and it reads as a move to align the "brain" of the coding agent on a single foundation — from the terminal all the way to the IDE.

One SDK, fewer hit-or-miss differences between tools

Until now, Copilot tended to behave and feel different depending on where it ran — CLI, VS Code, on GitHub, or in Visual Studio. This new agent is the first step toward consolidating those onto a single SDK. According to GitHub and Microsoft, the following changes distinguish it from the older agent.

• More likely to get it right the first time: designed to "handle the first exchange correctly and cut down on round trips."
• Shorter, easier-to-read responses: less time spent reading long explanations, so you can spend more time reviewing what it generates.
• Consistent across tools: aligning behavior across CLI, GitHub, VS Code, and Visual Studio.

Using it is simple: just pick "Agent (Preview)" from the agent selection menu at the bottom of Copilot Chat. For now, it is available on the Insiders channel of Visual Studio 2026.

Experts prepare and bundle the ".NET/Azure patterns"

The other pillar is built-in skills. Skills written by experts on the .NET and Azure teams appear in the "Built-in" section of the tool selection menu when you have the corresponding workload installed. The idea is to hand the agent established best practices as ready-made patterns, letting it work through a structured set of steps. Some of the published examples include the following.

The full list of skills is published in the dotnet/skills and microsoft/azure-skills repositories. They cover concrete development scenarios, such as creating ASP.NET Core endpoints and reviewing .NET performance.

Leaving it to "the user to choose" how they take effect

What stands out is that these skills ship off by default. GitHub assumes an approach where you "enable only the ones that fit the task in front of you, after reviewing them." Alongside this, in GitHub Copilot Business/Enterprise, organization owners can now set organization-level custom instructions that are applied automatically across repositories. It is a mechanism that lets the organization — not individuals — decide what the agent is allowed to do and which rules it runs under.

What to watch for when adopting it

From a practical standpoint, the safe move is to first try the new agent on an existing workflow and check whether the shorter responses have cut out context you actually need. Rather than enabling every skill, pick and turn on only the ones that help your team's work. If you use organization custom instructions, sorting out overlaps and conflicts with existing CLAUDE.md/AGENTS.md-style instructions ahead of time will help you avoid confusion later.

That said, there are downsides you shouldn't overlook behind the benefits. "Off by default" is a safety-oriented design, but the work of judging which skills are safe and useful still falls on the user. Even though the skills are made by experts, importing skills from external repositories means you can't skip the practice of verifying their source and what each update changes. And consolidating everything from the CLI to the IDE onto a single SDK, while it brings the benefit of consistency, also comes with a flip side: development teams grow more deeply dependent on one vendor's agent platform. It seems wise to view this in terms of both convenience and management cost.

References: GitHub Changelog: GitHub Copilot in Visual Studio — July update / Visual Studio Blog: Meet the new agent, powered by the Copilot SDK / Visual Studio Magazine: July Update Brings Copilot Agent and Built-In Skills / dotnet/skills / microsoft/azure-skills