AI Toolkit for VS Code: The .NET Developer’s Fast Track to Production Agents

TL;DR Microsoft released major updates to the AI Toolkit for VS Code designed to streamline how you build, test, and deploy AI agents . The biggest architectural shift is the transition from Copilot Instructions to Copilot Skills, which has equipped GitHub Copilot with specialized skills on developing AI agents using Microsoft Foundry and Agent Framework in a cost-efficient way . For .NET developers, this means you can now prototype, debug, and ship agentic workflows without leaving your IDE.

The Setup: Why This Matters Right Now

2026 is the year AI gets practical—the focus is shifting away from building ever-larger language models and toward the harder work of making AI usable, which involves deploying smaller models where they fit, embedding intelligence into physical devices, and designing systems that integrate cleanly into human workflows . Translation: demos are over. Shipping is in.

For .NET teams, the friction point has always been the gap between “I built a chatbot prototype” and “I deployed a production agent that reasons, calls APIs, and handles state.” This January update aims to close that gap by introducing powerful new debugging tools and enhancing support for enterprise-grade models via Microsoft Foundry .

What Changed: The Copilot Skills Migration

The transition from Copilot Instructions to Copilot Skills in v0.28.1 has migrated Copilot Tools from Custom Instructions to Agent Skills, allowing for more capable integration within GitHub Copilot Chat .

What this means for you:

Your AI Toolkit now understands agent workflows natively. Instead of generic instructions, you get specialized guidance on:

Orchestrating multi-step agent logic
Wiring up tools and APIs via Model Context Protocol (MCP)
Evaluating agent performance before shipping The custom AIAgentExpert now has a deeper understanding of workflow code generation and evaluation planning/execution .

Practical: Building an Agent in VS Code

Here’s the workflow you can use today:

1. Create an Agent Locally

Select the plus (+) icon next to the Agents subsection to create a new AI agent, after which both the agent .yaml file and the designer view open so that you can edit your AI agent .

A minimal agent YAML looks like:

name: DataProcessor
model: gpt-4
instructions: |
  You are an agent that processes CSV files and extracts insights.
  Use the file_search tool to read data, then summarize findings.
tools:
  - type: file_search
  - type: code_interpreter

2. Test in the Playground

You can now generate evaluation code directly within the toolkit to create and run evaluations in Microsoft Foundry . This means you can validate agent behavior before it touches production.

3. Deploy to Foundry Agent Service

By converging the Microsoft 365 Agents SDK with Microsoft Agent Framework—and aligning it with the shared runtime used in Foundry Agent Service—developers gain one unified set of abstractions to create, run, scale, and publish agents, so you can prototype locally, debug with consistent telemetry, and then seamlessly move into scaled hosting with enterprise-grade observability, compliance, and security .

The Cost & Latency Angle

The new Copilot Skills approach is cost-efficient , which matters because agent workflows can rack up token costs fast. The “bring your own model” feature allows enterprises to connect Foundry-hosted models through any AI gateway services like Azure API Management, Mulesoft, and Kong, to Foundry Agent Service, with the Agent Service honoring pre/post LLM hooks, policy-based model selection, and multi-region/multi-provider load-balancing with automatic failover, so teams can optimize for latency, cost, compliance, and availability without changing application code .

Translation: You can route expensive requests to cheaper models, cache results, and fail over regions—all declaratively—without rewriting your agent code.

Enterprise Wins: Anthropic + Azure

The Agent Builder and Playground now support Anthropic models using Entra Auth types, providing enterprise developers with a more secure way to leverage Claude models within the Agent Framework while maintaining strict authentication standards . If your org standardizes on Claude, you’re no longer locked into OpenAI endpoints.

The Bigger Picture: Unified Agent Development

The latest update to the VS Code AI Toolkit brings a streamlined experience for building with the Microsoft Agent Framework, enabling developers to locally create, run, and visualize multi-agent workflows, with enhancements that simplify the inner dev loop, making it easier to build, debug, and iterate on multi-agent systems within the familiar VS Code environment .

This is the missing piece for .NET teams. You no longer need to switch contexts between your IDE, a web portal, and a CLI. Everything lives in VS Code.

Next Steps

1. Install the AI Toolkit from the Visual Studio Code Marketplace . 2. Sign in at https://ai.azure.com and create a Foundry project, select a model (e.g., GPT-4.1 mini) and configure deployment options, and customize instructions to define your agent’s persona and tasks . 3. Test and iterate using the agent playground, then export code to Visual Studio Code for deployment .

The Reality Check

Be critical when vendors promise “80% accuracy” as if that’s the whole story—this is still generative AI in early 2026, so treat claims as marketing until you’ve seen working results in your own codebase, with your constraints, and your risk profile, and only believe what you can validate . The toolkit is solid, but your agents are only as good as the prompts, tools, and feedback loops you build around them.

AI Toolkit for VS Code: The .NET Developer's Fast Track to Production Agents