Foundry Toolkit for VS Code - Agents Workshop

April 2026 | Python, Microsoft Agent Framework, Azure OpenAI, Docker | GitHub | GA Announcement

On April 16, 2026, Microsoft announced the General Availability of Foundry Toolkit for VS Code - a complete AI development lifecycle extension bringing model exploration, agent prototyping, local debugging, and one-click production deployment directly into the editor. For the GA launch, I built and shipped the official hands-on workshop: two end-to-end labs that teach developers exactly how to use what was announced-from scaffolding a first agent to deploying a multi-agent system to Foundry Agent Service. The workshop was featured in the official GA announcement as the recommended training resource, and I am the sole author and maintainer of the project under the microsoft-foundry GitHub organization.

The GA Launch - What Was Announced

Foundry Toolkit GA shipped a unified in-editor experience for the full agent development lifecycle:

Model Playground - browse and test 100+ models from a single catalog without leaving VS Code
Agent Builder - visually prototype agents before writing any code
Agent Inspector - debug hosted agents locally with breakpoints, variable inspection, and live workflow visualization
Microsoft Agent Framework - Python SDK for building, customizing, and deploying hosted agents
Hosted Agent Deployment - one-click container build, push to ACR, and deploy to Foundry Agent Service
Foundry Playground - test deployed agents directly from the VS Code sidebar

These capabilities landed all at once on April 16. Developers had the tools. What was missing was a guided path to actually use them.

What I Built

I designed and built a two-lab workshop that maps directly onto every major GA capability-giving developers a structured, executable path from setup to a deployed multi-agent system. Each lab is self-contained, ships with a complete reference solution, and is organized into 9 step-by-step modules covering everything from prerequisites to troubleshooting.

Lab 01 teaches the fundamentals with a single agent. Lab 02 scales that foundation into a full multi-agent orchestration. Together, they cover the complete Foundry Toolkit lifecycle in under three hours.

Lab 01 - Single-Agent Fundamentals

“Explain Like I’m an Executive” Agent · ~45 min · 9 modules

Built a hosted agent that takes complex technical updates-incident reports, architecture changes, release notes-and produces plain-English executive summaries. The lab walks developers through every step of the GA toolkit lifecycle end-to-end:

Prerequisites and environment setup
Installing the Foundry Toolkit extension
Creating a Foundry project from VS Code
Scaffolding a hosted agent (auto-generated by the extension)
Configuring agent instructions via agent.yaml
Testing locally with Agent Inspector (F5 debugging, breakpoints, streaming output)
Deploying to Foundry Agent Service (Docker → ACR → Foundry, one click)
Verifying in Foundry Playground from the VS Code sidebar
Troubleshooting common failure modes

flowchart LR
    A([VS Code]) --> B[Scaffold Agent\nFoundry Toolkit]
    B --> C[Configure\nagent.yaml]
    C --> D[Test Locally\nAgent Inspector]
    D -->|pass| E[Deploy\nDocker → ACR → Foundry]
    D -->|fail| C
    E --> F[Verify\nFoundry Playground]
    F -->|issues| D
    F -->|✓| G([Production])

Reference solution: /agent/ - complete, runnable code with .env.example

Lab 02 - Multi-Agent Orchestration

“Resume → Job Fit Evaluator” · ~2–3 hours · 9 modules

Designed a four-agent collaborative system that demonstrates what production multi-agent development actually looks like on Foundry:

Agent	Role
Resume Parser	Extracts skills, experience, and education from uploaded resumes
Job Description Analyzer	Parses role requirements, must-haves, and preferred qualifications
Matching Agent	Scores candidate fit (0–100) against the parsed job requirements
Gap Analyzer	Produces a prioritized learning roadmap for missing skills

The lab covers multi-agent orchestration patterns (parallel execution, fan-out, sequential aggregation, conditional routing), inter-agent communication, and deploying a complex multi-agent workflow to Foundry Agent Service as a single unit.

flowchart TD
    U([User Input\nResume + Job Description]) --> RP
    U --> JA

    subgraph Parallel Parsing
        RP[Resume Parser\nSkills · Experience · Education]
        JA[Job Description Analyzer\nRequirements · Must-haves · Preferences]
    end

    RP --> MA
    JA --> MA

    MA[Matching Agent\nFit Score 0–100]

    MA --> GA[Gap Analyzer\nLearning Roadmap]

    GA --> O([Output\nScore + Prioritized Gaps])

Reference solution: /PersonalCareerCopilot/ - full orchestration implementation with deployment config

Tech Stack & Architecture

Languages: Python for agent logic, Docker for containerization, VS Code for development and deployment orchestration

Core SDK: Microsoft Agent Framework - agent scaffolding, agent.yaml configuration, hosted agent lifecycle management

Foundry Toolkit (VS Code Extension): auto-scaffolding, integrated Agent Inspector, deployment pipeline UI, Foundry Playground access from sidebar

Azure OpenAI: gpt-4.1 (Lab 02 orchestration), gpt-4.1-mini (Lab 01 single agent)

Deployment pipeline:

flowchart LR
    C[Agent Code\nPython + agent.yaml] --> D[Docker Build\nContainerize]
    D --> R[Azure Container Registry\nPush Image]
    R --> F[Foundry Agent Service\nDeploy + Run]
    F --> P[Foundry Playground\nTest from VS Code]

Configuration: modular .env per lab, no hardcoded credentials, .env.example included

Governance: Security Policy, Code of Conduct, Contribution Guidelines, CI workflows - all included from day one

Challenges Addressed

No guided onboarding existed at GA - Foundry Toolkit shipped with capability docs but no structured hands-on path. Designed the workshop to take a developer from zero to a deployed agent in under an hour.

Theory-to-practice gap - Official docs describe what each feature does. This workshop shows how to use them together in a real workflow. Every module is a practical step, not a concept explanation.

Full-stack surface area - Agent Inspector + Foundry Playground + Docker + ACR + agent.yaml is a lot to absorb at once. Structured the labs to introduce each layer at the moment it’s needed, so nothing feels arbitrary.

Cognitive leap from single to multi-agent - Most developers are comfortable with single-agent thinking. Lab 02 is explicitly designed to bridge that gap: same toolkit, same deployment path, four agents instead of one.

Impact & Recognition

Sole Maintainer - microsoft-foundry Organization I am the sole author and maintainer of this project within the official microsoft-foundry GitHub organization. Every module, reference solution, governance file, and CI workflow was built and is maintained by me. This is not a contribution to an existing project-it’s an official Microsoft Foundry resource I own end-to-end.

Featured in the Official GA Announcement The workshop was highlighted in the Foundry Toolkit GA announcement on April 16, 2026 as the recommended hands-on training resource for developers adopting the platform. Being featured at launch-not added as an afterthought-means it was recognized as launch-ready quality by the Foundry team.

Extends Across the GA Ecosystem The April 16 GA event announced capabilities beyond the Foundry Toolkit extension-including updates to Foundry Agent Service, the Microsoft Agent Framework SDK, and broader Azure AI Foundry platform features. This workshop is designed to grow with those announcements. As new GA capabilities land, labs will be extended to cover them-keeping the workshop current and positioning it as the living, community-facing onboarding path for the entire Foundry GA wave, not just the VS Code extension.

Learning Outcomes

Shipped an official project under the microsoft-foundry GitHub organization as sole author and maintainer
Designed a two-lab progressive curriculum (single-agent → multi-agent) that maps directly to GA launch capabilities
Built and validated the complete Foundry Toolkit lifecycle: scaffolding → local testing → deployment → cloud verification
Implemented four multi-agent orchestration patterns: parallel execution, fan-out, sequential aggregation, and conditional routing
Practiced production-grade containerization: Docker image build, ACR push, and Foundry Agent Service deployment
Authored developer-facing documentation across 18 step-by-step modules with troubleshooting guides and reference solutions
Featured in the official Foundry Toolkit GA announcement as the recommended hands-on training resource for the launch

Foundry Toolkit for VS Code - Agents Workshop (Microsoft Foundry)