Implement generative AI apps and agents by using Foundry — AI-901 Study Guide

Unlock the complete study guide + 1,040 practice questions across 16 full exams.

Bundled into the existing Microsoft Azure AI Fundamentals premium course — no separate purchase.

14-day money-back guarantee — no questions asked.

Included in this chapter:

From a deployed model to a working application
What belongs in the system prompt
Sharpening a prompt: specifics, examples, and an out
Deploy a model and test it in the Foundry portal
A lightweight chat client with the Foundry SDK
What a prompt agent is made of
Testing an agent in the agents playground
Controlling whether a tool gets called
Calling an agent from application code
How AI-901 phrases questions about this objective

Where the configuration lives for the same request

Comparison point	Foundry playground	Model client (Foundry SDK)	Agent client (Foundry SDK)
Instructions and tools are set	In the portal, for the session you are testing	In each request your code sends	In the stored agent definition
The request identifies	The deployment whose playground is open	The deployment name	The agent name, through an agent reference
Code you write	None	Python client code	Python client code
Conversation history comes from	The playground conversation	A conversation you create, or a transcript your code resends	A conversation you create, or the previous response
Changing the behavior means	Editing the form and retesting	Editing and redeploying your application	Creating a new version of the agent definition
Reach for it when	Iterating on prompts and first checks	One application needs a generation call	Several applications share one configured behavior

Decision tree

Cheat sheet

Cheat sheet

System prompts set high-priority model behavior
User prompts supply the current task and content
Specific instructions constrain model output
Few-shot examples demonstrate the desired output pattern
Fallback instructions limit unsupported answers
A model must be deployed before ordinary playground inference
The playground Code tab exposes client access details
The Foundry portal unifies model discovery, deployment, and testing
The model parameter uses the deployment name
The Foundry SDK project client opens the model client
responses.create sends text input and exposes output_text
An agent reasons with a model, instructions, and tools
Prompt agents are managed as named, versioned definitions
The Agents playground tests conversation and tool behavior
Agent clients invoke a server-side definition by agent reference
Required tool choice forces an available tool call

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References