Copilot Chat: Build Declarative Agents

Key insights

• Coding agents: AI tools that write, debug, and assemble the files a declarative agent needs (TypeSpec, JSON manifest, instructions) from a natural-language description, cutting development time and manual errors.
  
• Declarative agents: custom Copilot experiences declared by configuration rather than custom orchestrators; they run on Copilot’s orchestrator and foundation model so you don’t need extra hosting and they align with Microsoft 365 security and compliance.
  
• Agent Toolkit (ATK): scaffolds agent skills, wires capabilities (APIs, connectors, Teams messages, code interpreter), validates setup, and helps deploy an agent from an empty folder, as shown in the demo by Sébastien Levert.
  
• Key advantages: faster build cycles, lower technical barriers, built-in governance controls (disclaimers and behavior overrides), flexible knowledge connectors, and smoother enterprise integration with APIs and Teams.
  
• Typical workflow: describe the agent in plain language → coding agent generates configuration and code → test and validate behavior → deploy the declarative agent into Copilot; this reduces repetitive manual configuration steps.
  
• Best practices: review generated artifacts, add clear instructions and RAI-compliant safeguards, verify connectors and indexing scopes, and test with representative Teams conversations before production use.
  

Microsoft released a practical demo that shows how coding agents can accelerate the creation of Copilot Chat declarative agents, and the demonstration highlights a fast, repeatable workflow for developers. The video, presented by Sébastien Levert during a Microsoft 365 & Power Platform community call, walks viewers through building an agent from an empty folder to deployment. Therefore, the session emphasizes automation, reproducibility, and direct integration with Microsoft 365 capabilities that many teams already use. Consequently, the demo frames a clear shift from hand-coding manifests toward a more narrative-driven development process.

Demo at a glance

The presenter begins by showing how an Agent Toolkit skill can scaffold the core files such as the manifest, instruction files, and TypeSpec artifacts that declarative agents require. Next, the demo uses conversational prompts taken from a Teams exchange to derive requirements and to generate structured instructions automatically. Then, the coding agent wires up capabilities like Teams messages, API calls, and custom connectors so the agent can interact with enterprise systems in real time. Finally, the video demonstrates validation steps and a one-step deploy that moves the agent from a local folder into a runnable Copilot extension.

How coding agents scaffold agents

First, the coding agent interprets plain-language requirements and turns them into technical artifacts, which reduces manual errors and speeds initial setup. Moreover, the agent generates the JSON manifests and TypeSpec files that define endpoints, permissions, and conversational behavior without deep schema knowledge from the developer. In addition, the process includes checks and validation so developers can catch misconfigurations before deployment. As a result, teams gain a repeatable way to produce working declarative agents while maintaining consistency across multiple projects.

Benefits and practical tradeoffs

The workflow delivers clear benefits such as faster time to prototype, lower barrier to entry, and better alignment with Microsoft 365 governance defaults. However, speed comes with tradeoffs: although coding agents automate repetitive tasks, they may produce generic patterns that need refinement for complex or regulated scenarios. Therefore, teams should expect to invest time in tailoring generated instructions, testing edge cases, and encoding organizational policies into the templates. In short, automation improves productivity but does not eliminate the need for human oversight and design tradeoffs.

Implementation challenges and governance

Security and data governance remain central challenges when connecting Copilot to enterprise content and APIs, and the demo points out that connectors and indexing must be configured carefully. Furthermore, organizations will need robust validation, logging, and user-facing disclaimers so that the agent’s behavior meets compliance and Responsible AI requirements. At the same time, automated scaffolding can embed governance controls early in the lifecycle, which reduces drift and inconsistent policies across agents. Nevertheless, teams must balance the convenience of automation with disciplined review processes and continuous monitoring.

Developer and organizational implications

For developers, the approach lowers the technical threshold and encourages iteration by turning natural language design into deployable artifacts quickly. Meanwhile, architects and security teams gain opportunities to standardize templates and approvals so that generated agents conform to corporate controls. Consequently, organizations that adopt this method can scale prototypes into production with less friction, provided they maintain a clear testing and governance workflow. Ultimately, the balance between speed, control, and safety will determine how broadly teams apply this model.

Conclusion and next steps

The demo showcased by Microsoft and presented by Sébastien Levert highlights a practical path to building declarative agents using coding agents, and it demonstrates concrete benefits for prototyping and consistent delivery. That said, teams should weigh the tradeoffs between automation and bespoke customization, and they should plan governance and validation steps before broad deployment. Therefore, organizations curious about this workflow will find value in experimenting with the toolkit while documenting policy guardrails and test scenarios. In conclusion, the demo points toward faster agent development, but success still depends on sensible oversight and iterative refinement.

Keywords

copilot chat declarative agent, coding agents for copilot, build copilot chat agent, declarative agent tutorial, coding agents examples, microsoft copilot agent guide, autonomous coding agents, copilot agent best practices