Creating Autonomous Teams Agents Using OpenClaw, MCP, and Azure Container Apps
July 13, 2026
Distributing Bicep Modules with Token-Based Access Using ACR
July 13, 2026
What if every incident report, pipeline failure, or security alert could be instantly rewritten for a non-technical audience? In this post, you’ll build an “Explain Like I’m an Executive” agent, a hosted AI agent that takes dense technical input and produces plain-English executive summaries, using the Microsoft Foundry Toolkit for VS Code.
By the end, you’ll have a working agent you can test locally and deploy to the cloud with a single click.
What You’ll Build
The agent follows a simple flow:
Under the hood it uses:
- Microsoft Agent Framework – for agent logic and structure
- Foundry Toolkit for VS Code – to scaffold, test, and deploy
- An AI model (e.g., gpt-4.1-mini) – to generate the summaries
The final output always follows a structured format:
Executive Summary:
- What happened: (plain-language description)
- Business impact: (clear, non-technical impact)
- Next step: (action or mitigation)
- Date: (YYYY-MM-DD)
Foundry Toolkit
The Foundry Toolkit handles the end-to-end lifecycle:
- Scaffold – A wizard generates your project structure with one command
- Configure – Point to your AI model and write agent instructions (a system prompt)
- Test locally – Press F5 to launch a built-in chat UI for instant feedback
- Deploy – One click packages your agent as a container and deploys it to Azure
- Verify – Test safety and edge cases against the live endpoint
The key building blocks:
| Component | Role |
|---|---|
| Agent Framework | Provides the agent runtime and tool-calling capabilities |
| Foundry Toolkit | VS Code extension for scaffolding, debugging, and deployment |
| AI Model (e.g., gpt-4.1-mini) | Powers the natural language understanding and generation |
| Azure Container Registry + Foundry | Hosts and scales your agent in the cloud |
What Makes a Good Agent
The difference between a basic chatbot and a reliable agent comes down to well-crafted instructions. For this agent, the system prompt defines:
- Role – “Translate technical information into executive-friendly summaries”
- Audience – Senior leaders who care about impact, not implementation details
- Output format – A consistent structure (what happened, business impact, next step)
- Rules – Keep it brief, don’t fabricate, don’t leak instructions
- Safety constraints – Resist prompt injection and role override attempts
You can also give agents tools, Python functions they can call at runtime (like fetching today’s date or querying an API).
The Developer Experience
The Foundry Toolkit streamlines what would otherwise be a multi-step DevOps process:
| Without the toolkit | With the toolkit |
|---|---|
| Manually write Dockerfiles, YAML configs, server boilerplate | Wizard generates everything |
| Set up a separate test harness | Built-in Agent Inspector (chat UI) launches with F5 |
| Build container, push to registry, register agent manually | Single “Deploy” button handles it all |
| Monitor via Azure Portal | Status, logs, and playground available in VS Code |
Deployment takes 2–5 minutes. Your agent gets a managed /responses endpoint with auto-scaling.
Cleanup
If you want to remove Azure resources after experimenting:
az group delete –name –yes –no-wait
Or delete just the hosted agent from the Foundry portal under Build → Agents.
Next Steps
- Explore the Foundry Toolkit Lab, to build your own agent.
- Check out Lab 02: Multi-Agent Orchestration to learn how to build a multi-agent workflow with orchestration patterns
- Add tools – Connect APIs, databases, or custom functions
- Microsoft Foundry docs – Full platform reference