Copilot Chat: Add Your Custom Agent

Key insights

• Custom engine agents demo: Ayça Baş (Microsoft) shows how to make an agent appear in Microsoft 365 Copilot Chat by updating the agent manifest, moving to schema 1.22, and adding Copilot agent tags.
  She tested the agent in Teams and the Playground to confirm behavior across environments.
• Definition and control: Custom engine agents are specialized AI assistants that give developers full control over orchestration, models, and data integrations, unlike fixed declarative agents.
  This lets teams build workflows tailored to their business needs instead of fitting into preset templates.
• Cost and licensing: these agents use a pay-per-resource approach rather than requiring Copilot licenses for every user, improving cost efficiency and lowering total cost of ownership for large deployments.
  Organizations can scale agents without the same per-user licensing expense.
• Flexibility and reach: developers can choose foundation or fine-tuned models, connect to external systems, and deploy agents across Copilot, Teams, web apps, and custom platforms for true multi-channel coverage.
  This preserves existing infrastructure choices and supports diverse user touchpoints.
• Automation and proactivity: custom agents can run long tasks in the background and send proactive alerts, enabling background automation and more advanced workflows than simple request-response models.
  That helps keep users informed and reduces manual steps.
• Developer tools and quick start: Microsoft provides the Microsoft 365 Agents SDK (including NodeJS/TypeScript templates), integration with Visual Studio/VS Code, and support from the Semantic Kernel to speed development.
  Updating the manifest to the correct schema and adding Copilot tags is the key step to appear in Copilot Chat.

Introduction

The Microsoft-produced YouTube demo showcases how developers can bring a Custom Engine Agent into Microsoft 365 Copilot Chat with a small manifest update. Presented by Ayça Baş during a Microsoft 365 & Power Platform community call, the video walks through a working agent built with the Microsoft 365 Agents SDK and Semantic Kernel, tested in Teams and the developer playground. Moreover, the presenter demonstrates that updating the manifest to schema 1.22 and adding the Copilot agent tags makes the agent appear inside Copilot Chat.

This story summarizes the demo objectively and explains what the integration means for enterprise AI deployments. In particular, it highlights practical steps and consequences rather than promotional language. Consequently, editors can use this piece to inform readers about both the capabilities and the realities of adopting custom agents in Microsoft 365.

How the Integration Works

First, the demo shows a concrete, hands-on approach: developers build an agent using the Microsoft 365 Agents SDK and the Semantic Kernel to manage orchestration and reasoning. Then, by conforming to the newer manifest schema 1.22 and tagging the agent for Copilot, the same runtime can be surfaced inside Microsoft 365 Copilot Chat. Therefore, teams can avoid rebuilding parallel experiences for Copilot and other channels.

In addition, the video clarifies that the agent architecture supports multi-channel deployment. For example, the agent tested in Teams and a local playground appears in Copilot Chat after the manifest tweak, which demonstrates consistent behavior across endpoints. Consequently, organizations can maintain a single codebase while reaching users in different interfaces.

Finally, the presenter briefly notes that the integration relies on clear schema compatibility and tag conventions. Thus, upgrading agents to the correct schema and validating tags are practical prerequisites before release. As a result, development and release processes must account for manifest versioning and deployment checks.

Business Benefits

Custom engine agents offer tangible operational advantages for many enterprises, beginning with greater control over orchestration and model choice. For example, teams can pick foundation models, deploy fine-tuned variants, or route specific requests to specialized services; therefore, they can align AI outputs with business rules and compliance needs. Additionally, the pay-per-resource cost model associated with custom agents can reduce licensing overhead compared with granting Copilot seats to every user.

Moreover, the multi-channel capability extends the agent’s value beyond a single app, which increases reach and adoption without duplicating engineering effort. In practice, this means the same agent can automate tasks inside email, chat, and custom portals, while also running background workflows. Consequently, organizations can deliver proactive alerts and long-running processes that keep users informed without constant manual input.

Lastly, the flexibility of custom agents supports complex enterprise scenarios that declarative or studio-driven agents may not accommodate. For instance, agents can orchestrate cross-application actions and integrate with legacy systems. However, unlocking this flexibility demands engineering investment and appropriate governance.

Tradeoffs and Challenges

Despite the advantages, the demo implicitly highlights several tradeoffs that teams must weigh. On one hand, custom agents provide deep control and cost benefits; on the other hand, they increase responsibility for security, model selection, and ongoing maintenance. Therefore, organizations must invest in monitoring, access controls, and testing to ensure the agent behaves safely at scale.

Moreover, complexity grows when integrating with a variety of data sources and enterprise applications. For example, connecting to internal services may require secure connectors, throttling strategies, and data sanitization to avoid leaks or performance issues. Consequently, governance frameworks and clear ownership become essential to manage operational risk and compliance obligations.

Finally, the need to manage schema versions and compatibility introduces another operational burden. While updating a manifest to schema 1.22 may be straightforward in many cases, teams should plan for regression testing and backward compatibility. Thus, balancing rapid innovation with stable production behavior remains a central challenge for adopters.

Development Toolkit and Next Steps

The video underscores practical developer tools: the Microsoft 365 Agents SDK and the associated toolkit streamline agent creation, debugging, and local testing. For example, templates and IDE integrations help reduce boilerplate work and speed up the path to a working prototype. Consequently, teams can iterate faster while validating behavior in both Teams and Copilot Chat environments.

Furthermore, the presenter demonstrates a useful development pattern: validate agents in a controlled playground and then enable them in Copilot through manifest changes. This approach encourages safe rollouts and staged testing before broad deployment. Therefore, organizations should adopt similarly cautious release practices, including staged user testing and telemetry-driven rollouts.

In summary, the demo provides a clear, practical path for bringing custom agents into Copilot Chat while also highlighting the operational work required. As enterprises evaluate this approach, they should balance the benefits of control and cost-efficiency against the demands of security, governance, and ongoing maintenance.

Keywords

Microsoft 365 Copilot custom agent, Copilot Chat custom engine integration, Bring custom agent to Copilot Chat, Custom AI agent for Microsoft 365, Copilot extensibility and custom engines, Build custom Copilot Chat agent, Copilot SDK custom agent tutorial, Enterprise Copilot custom engine deployment