Foundry agent basics

Guidance for building a first agent in this course (Lessons 1–2).

Guardrails (always apply)

• Platform: Microsoft Foundry (never say "Azure AI Foundry"). Framework: Microsoft Agent Framework (MAF).
• SDK: use the pinned agent-framework from requirements.txt with FoundryChatClient and the Responses API. Do not use AzureAIClient, AzureAIAgentClient, or AzureOpenAIChatClient.
• Model: gpt-5.1 for chat. Never gpt-4o / gpt-4.1 (retired) and never GitHub Models.
• Auth: az login + AzureCliCredential; endpoint/model come from .env (never hardcoded).

What a learner should understand first

1. Agent = model + instructions + tools + a thread. Instructions set the persona/task; tools extend capability; a thread carries conversation state.
2. Design before code (Lesson 1): identify the scenario, the user, the tools needed, and the success criteria. The course use case is "Developer Onboarding".
3. One agent, one job. Lesson 2 builds several small, specialised agents rather than one monolith (employee search, learning recommendation, coding, orchestration, etc.).

Minimal shape of a sample

• Load env with python-dotenv; read the Foundry project endpoint and model name.
• Create a FoundryChatClient with AzureCliCredential.
• Create the agent with clear instructions; attach tools if needed.
• Run it — most lesson-2 samples expose a local DevUI on a dedicated port (8090+).

How to help

• Point the learner at the closest existing sample in lesson-2-agent-development/ and adapt it, rather than writing from scratch.
• Keep instructions specific and scoped; show how to test with a couple of representative prompts.
• Remind them to pip install -r requirements.txt, az login, and populate .env first.

Validate

• python -m py_compile <file>.py must pass.
• The sample should start its DevUI and answer a basic prompt end-to-end.

References

• lesson-1-agent-design/README.md, lesson-2-agent-development/README.md
• MIGRATION-GUIDE.md for the current SDK patterns.