Vibe Code a Foundry Hosted Agent using VS Code and GitHub Copilot

Author - Kunal Saha 🔸Azure Solution Architect🔸Azure Integration Specialist

Following the Build 2026 announcement, I was quite keen to explore the Microsoft Foundry Toolkit for VS Code, which is now generally available.

You can read more about it here: Foundry Toolkit for Visual Studio Code

So, today’s blog is focused on sharing my findings from vibe coding a Foundry agent and deploying it as a hosted agent.

Before we dive further, let’s quickly understand the difference between Prompt Agents and Hosted Agents in Microsoft Foundry:

“Hosted agents are containerized agentic AI applications that run on Agent Service. Unlike prompt-based agents — which are defined entirely through prompts and tool configuration in the Foundry portal — Hosted agents are your own code packaged as a container image. You choose the framework, control the runtime behavior, and deploy the image to Microsoft-managed infrastructure.”

🎯 What are we building ?

So now, coming to the actual agenda — today we are vibe coding a Writer–Reviewer Multi-Agent Workflow.

This is a multi-agent workflow application that demonstrates content collaboration using the Microsoft Agent Framework, it implements a Writer-Reviewer pattern where:

1. Writer Agent creates initial content based on user input

2. Reviewer Agent provides feedback and refinement

3. Workflow Output is the refined content after collaboration

Let's talk about some of the Prerequisites

• Python 3.11+
• Azure AI Foundry resource with a project and a deployed LLM
• VS Code with the Foundry Toolkit extension
• Toolkit configured with your Foundry resource

Step-by-Step walkthrough

Alright, let’s do this step by step.

Step 1 — Create the agent using GitHub Copilot

Use the following prompt to initiate the vibe-coding session:

“Create a multi-agent workflow app with Agent Framework SDK. Scenario: Writer-Reviewer content collaboration — A multi-agent workflow where a Writer and Reviewer collaborate to create and refine content. High-level flow: 1) writer agent receives a user request and creates initial content. 2) The content is passed to reviewer agent, who provides concise, actionable feedback. 3) Both the writer and reviewer are output executors, so the workflow yields their collaborative result as the final output. Output: The final workflow output is plain text containing the refined content after the writer-reviewer collaboration. “

Step 2— Let Copilot do its thing

Wait for GitHub Copilot to perform all the necessary steps. Allow actions wherever required.

Copilot will:

• Provide an implementation plan
• Generate all required project files
• Create supporting markdown documentation

Step 3 — Debug locally

Run the agent in VS Code:

Run → Start Debugging

Step 4 — Test locally

Use the Agent Inspector Playground with a sample prompt like:

“Write a technical blog about AI safety considerations”

PS — Don’t forget to update your .env file so it connects to your Foundry project and model.

Step 5 — Deploy as a Foundry Hosted Agent

Click the Deploy button in the Agent Inspector window.

You’ll be able to:

• Select deployment method
• Choose container registry
• Provide a meaningful name

In my case, I selected:

• Deployment Method: Docker Image via ACR

This creates a new ACR resource in the same resource group as the Foundry resource.

Wait for deployment to complete successfully.

Step 6— Verify the hosted agent on the Microsoft foundry platform

Login to Microsoft Foundry → Go to Agents and boom, your agent is there 🙂

Step 7 — Test the hosted agent

Now give it a fresh prompt to generate new content

Wait for the agent to respond with another brand-new technical blog

Step 8 — Integrate with client apps

You can now integrate this agent with another application using the Python SDK.

Final Thoughts

This end-to-end experience shows how GitHub Copilot and the Foundry Toolkit can really accelerate building and deploying agentic AI workflows.

The ability to:

• Rapidly scaffold multi-agent patterns
• Test locally with rich tooling
• Deploy as production-grade hosted agents

makes this combination a strong addition to the Microsoft AI ecosystem.

It brings together speed, simplicity, and production readiness — helping teams move from idea to working intelligent systems much faster, without compromising on quality or scalability.