The Microsoft Foundry Toolkit extension for Visual Studio Code integrates AI model management, agent building, evaluation, and deployment into a unified development environment, enhanced by GitHub Copilot for AI-assisted workflows. It supports multi-model ecosystems, low-code and pro-code agent creation, interactive testing, and seamless cloud deployment, enabling developers to efficiently build, refine, and govern AI applications within VS Code.
The video introduces the Microsoft Foundry Toolkit extension for Visual Studio Code, designed to streamline the development of AI applications and agents by integrating models, prompts, evaluation, and deployment within the editor. Microsoft Foundry serves as a unified AI platform that supports building, grounding, evaluating, deploying, and governing AI applications at scale, accessible via a web portal and SDKs. The Foundry Toolkit extension, combined with Microsoft Foundry skills integrated into GitHub Copilot, allows developers to seamlessly access and manage AI models, agents, and workflows directly from Visual Studio Code, eliminating the need to switch between multiple tools.
The extension is organized into three main sections: My Resources, Developer Tools, and Feedback. My Resources provides access to recent agents, local resources, Microsoft Foundry cloud resources, and connected external resources. Developer Tools include discovery features like the model and tool catalogs, build tools for creating and debugging agents, deployment capabilities, and telemetry for performance analysis and evaluation. The integration with GitHub Copilot enhances productivity by offering AI-powered assistance throughout the development workflow, including project setup and agent creation, all within the Visual Studio Code environment.
A key feature demonstrated is the model catalog and playground, which help developers explore, compare, and select the most suitable AI models for their scenarios. The toolkit supports models from multiple providers such as Microsoft Foundry, OpenAI, Anthropic, and local models, enabling a flexible multi-model ecosystem. Developers can use GitHub Copilot to get model recommendations based on specific criteria like deployment region and capabilities, then validate and deploy chosen models directly within the toolkit. The playground allows side-by-side comparison of models, testing prompts, and analyzing responses to make informed decisions.
The video also showcases building AI agents using both low-code and pro-code approaches. Using the agent builder, developers can create prompt-based agents with clear instructions, integrate tools like the Microsoft Learn MCP server for accessing official documentation, and test agents interactively. The toolkit supports AI-assisted evaluation with metrics such as task adherence and groundedness, enabling iterative refinement of agents before production deployment. This process ensures agents are reliable, grounded in validated data, and tailored to specific business needs, all managed within Visual Studio Code.
Finally, the video covers advanced agent development using GitHub Copilot CLI and Microsoft Foundry skills to generate coded agents deployable as hosted agents in Microsoft Foundry. The generated projects include configuration files, toolboxes, and deployment templates, facilitating integration with existing workflows and source control. Developers can debug agents locally using the agent inspector, monitor tool usage and traces, and deploy agents to the cloud with ease. This comprehensive workflow empowers teams to build, test, and deploy sophisticated AI agents efficiently while maintaining transparency and control throughout the development lifecycle.