Copilot Studio agent building with coding assistants

Key insights

• The demo covers a Copilot Studio plugin workflow with subagents for advising, authoring, managing and testing.
• Plan mode is included in the demonstrated workflow.
• The Copilot Studio VS Code extension is used to clone an agent locally.
• The presenter demonstrates YAML validation and pushing changes back to Copilot Studio.
• The demo also covers adding SharePoint knowledge and generating adaptive cards from the command line.

Microsoft published a 17-minute demo showing how to build Copilot Studio agents with GitHub Copilot or Claude Code. The video description says it was presented during a Power Platform Monthly Community call, but the supplied page metadata does not state when the video was published. The main takeaway is that the demonstrated workflow connects coding assistants with Copilot Studio authoring tasks, including local agent work, validation and testing.

The description identifies several components but does not provide a full transcript or a step-by-step technical guide. This article therefore focuses on what the video description says is demonstrated, rather than treating the listed activities as a complete product specification. For makers and administrators, the useful distinction is between those concrete workflow examples and details that remain unreported, such as licensing, prerequisites and availability.

What's new

The demo brings together a Copilot Studio plugin workflow and the Copilot Studio extension for Visual Studio Code. Its description names subagents for advising, authoring, managing and testing agents, as well as a plan mode. It also says the extension is used to clone an agent locally, validate YAML and push changes back to Copilot Studio.

Other activities named in the description are adding SharePoint knowledge and generating adaptive cards from the command line. Taken together, the listed examples span planning, agent authoring, local file work, validation and returning changes to the service. They do not establish that every agent uses all of these steps, or that all of the work happens automatically.

The title identifies GitHub Copilot and Claude Code as options for the demonstrated agent-building approach. The description does not say whether the two coding assistants have identical capabilities in this workflow, or explain which assistant performs each action. It therefore supports a comparison of the approach at a high level, but not a detailed feature-by-feature comparison.

Why it matters

For makers and developers, the video is relevant because it shows an agent workflow that includes both Copilot Studio and local development tools. Cloning an agent locally and validating YAML are the most concrete examples of work outside the Copilot Studio authoring surface. The description also includes testing among the subagent responsibilities, indicating that the workflow covers more than writing agent content.

For teams, the described sequence may help frame questions about how agent changes are prepared and checked before they are pushed back to Copilot Studio. The source does not say that the workflow replaces the Copilot Studio interface, nor does it specify a required development process. It is best understood as a demonstration of tasks that can be brought together, not a documented mandate or complete deployment method.

Administrators may notice the reference to adding SharePoint knowledge, but the description does not explain how SharePoint content is selected, permission-checked or maintained. Those operational details matter when an agent uses organizational information, yet they cannot be inferred from this brief description. End users are not the primary audience for the demonstrated steps: the account is about building and managing agents rather than a change to how users interact with them.

The video gives no licensing, rollout or availability information. It does not name prerequisites for GitHub Copilot, Claude Code, the VS Code extension or Copilot Studio, and it does not state whether the demo reflects a preview feature or a generally available capability. Organizations should not read the video description as confirmation that the workflow is enabled in every environment.

How it works

The description outlines a sequence of activities, though not enough detail to reproduce the setup. It names the plugin workflow, several subagent roles and plan mode, then identifies local agent work and a return path to Copilot Studio. The following list reflects only those actions named in the supplied material:

1. Use the Copilot Studio plugin workflow with GitHub Copilot or Claude Code.
2. Use subagents described as advising, authoring, managing and testing agents, with plan mode also included.
3. Use the Copilot Studio VS Code extension to clone an agent locally.
4. Validate YAML and push changes back to Copilot Studio.
5. Add SharePoint knowledge and generate adaptive cards from the command line.

These points are a summary of the demo description, not a verified set of required steps in a prescribed order. In particular, the supplied material does not detail how the agent is cloned, what YAML is validated, or what happens if validation fails. It also does not explain how the command-line adaptive card generation relates to the other stages.

The mention of subagents is similarly brief. The description assigns them broad roles—advising, authoring, managing and testing—but does not explain their configuration, boundaries or interaction. Plan mode is named without a description of its interface or output. Readers looking for commands, prompts, configuration examples or troubleshooting guidance will need more technical material than the video description provides.

Limitations and open questions

The source available for this article is a short video description rather than a transcript or technical guide. It identifies workflow components but leaves details such as prerequisites, supported agent types, authentication, error handling and testing criteria unspecified. The description also does not say whether the demonstrated tools or workflow are in preview, generally available or subject to particular licensing.

Although the description links to the Copilot Studio VS Code extension, it does not establish its licensing requirements or availability in any particular tenant. Nor does it provide enough detail to assess how SharePoint knowledge is governed or how changes are reviewed before being pushed back. The video is useful as an overview of the demonstrated activities, but readers should seek product documentation for implementation details.

We also cannot verify from the supplied description how the GitHub Copilot and Claude Code experiences differ, or whether every task shown applies to both. No claims about costs, supported regions, version requirements or general rollout are warranted from the information provided. Those gaps limit how far the demonstration can support an adoption decision on its own.

Sources

• Microsoft video: Building Copilot Studio Agents with GitHub Copilot or Claude Code

Frequently asked questions

What tools does the demo use?

The description names GitHub Copilot or Claude Code, the Copilot Studio plugin workflow and the Copilot Studio VS Code extension.

What does the workflow demonstrate?

It covers agent subagents for advising, authoring, managing and testing, plan mode, local cloning, YAML validation and pushing changes back to Copilot Studio.

Does the video specify licensing or availability?

No. The supplied description does not state licensing requirements, rollout status or availability.