AI-103 · Practice set 3 of 13
Foundry Agent Service at a Glance: 10 practice questions
10 free AI-103 practice questions on Foundry Agent Service at a Glance, with an explanation for every answer. Untimed. The full mock exam and the timed version are in the app.
-
Question 1 of 10
For a text-based internal tool, which agent type is the fastest path, defined only by instructions, a model, and tools, with nothing to host?
- AAn agent packaged as a .zip file
- BA prompt agent
- CA hosted agent
- DA voice-based prompt agent
Show the answer
For a text helper with no custom runtime code, choose the configuration-only type: instructions, model, and tools are enough, and Foundry hosts it. The hosted type is for code you bring; the voice type is for spoken interaction.
Next → 1 / 10 -
Question 2 of 10
What does a toolbox provide in Foundry Agent Service?
- AAutomatically improved instructions and tool descriptions
- BEnd-to-end tracing of every decision the agent makes
- CHosting and scaling for prompt agents and hosted agents
- DCurated tools behind one managed MCP-compatible endpoint
Show the answer
Each component has one job. A toolbox is about tools: you curate them once and expose them through one managed MCP-compatible endpoint, with authentication, governance, and versioning handled in a single place.
Next → 2 / 10 -
Question 3 of 10
Agent Service works with models from the Foundry catalog, such as GPT-4o, Llama, and DeepSeek. What does it let you do with an agent's model?
- AUse only models from Microsoft and OpenAI
- BSwap it without changing the agent code
- CRewrite the agent code whenever it changes
- DManage its compute yourself for prompt agents
Show the answer
The model is a swappable part. Because the service fronts many catalog models through one entry point, you can change which model an agent uses without touching the agent code.
Next → 3 / 10 -
Question 4 of 10
A team has its hosted agent as source code and no container image. What happens if it ships the code as a .zip file?
- AFoundry runs it as a voice agent over a WebSocket
- BFoundry builds the container image for you
- CFoundry rejects it and asks for a container image
- DFoundry converts it into a prompt agent configuration
Show the answer
Hosted agents always run as containers, but you do not have to build the image yourself. You can ship a container image or a .zip of your source, and in the .zip case Foundry builds the image.
Next → 4 / 10 -
Question 5 of 10
An agent calls a remote MCP server that must act with the signed-in user's own permissions. Which authentication option fits?
- AUnauthenticated access to the server
- BMicrosoft Entra with the project's managed identity
- CKey-based access with a shared key
- DOAuth identity passthrough (On-Behalf-Of)
Show the answer
Ask whose identity the server should see. On-Behalf-Of passes the user's identity through, so the server applies that user's permissions. Keys and managed identities authenticate the agent or the project instead.
Next → 5 / 10 -
Keep the ones you got wrong
In the app, every question you miss comes back exactly when you’re about to forget it.
-
Question 6 of 10
Which cost does a hosted agent add that a prompt agent does not have?
- ATool usage
- BVoice Live usage
- CPer-call model inference
- DContainer compute
Show the answer
Compare what both types pay for first: inference calls and tool use. The code-based type also runs your code in a managed container, and running that container adds a separate charge.
Next → 6 / 10 -
Question 7 of 10
An agent must call an internal inventory API that is described by an OpenAPI spec. Which kind of tool does this need?
- AThe built-in code interpreter tool
- BThe built-in web search tool
- CA custom tool built from the OpenAPI spec
- DThe built-in file search tool
Show the answer
Built-in tools cover general jobs such as searching the web or files and running code. Your own API is specific to you, so you add it as a custom tool, and an OpenAPI spec is one of the ways to do that, alongside functions and MCP servers.
Next → 7 / 10 -
Question 8 of 10
A team wants its agent's instructions, tools, and model to live in the app's own repository, reviewed in pull requests, with no agent resource to create or update in Foundry. It still wants Foundry models and tools. Which approach fits?
- ADefine a prompt agent code-first through the SDK
- BPackage the agent as a hosted agent container
- CPublish the agent to get a stable endpoint
- DCall the Responses API directly from the app
Show the answer
Two constraints decide it: the definition must live in app code, and there must be no Foundry agent resource. Calling the Responses API directly builds an ephemeral agent that is assembled in your process on each call, while still using Foundry models and tools.
Next → 8 / 10 -
Question 9 of 10
A contact center voice agent must use a specific Azure voice and a phrase list for product names, with explicit control over transcription. The team accepts some extra latency. Which voice architecture fits?
- AA cascaded text model
- BA text prompt agent with a toolbox
- CA hosted agent with webhooks
- DNative speech-to-speech
Show the answer
Weigh control against latency. Native speech-to-speech is the low-latency option, but the cascaded chain of speech recognition, text model, and speech synthesis is what offers broader Azure voice choices, phrase lists, and explicit transcription control, at the cost of latency per stage.
Next → 9 / 10 -
Question 10 of 10
An agent must call into the team's own Python library and use custom orchestration built with LangGraph. The team wants Foundry to handle the endpoint, scaling, and identity. Which approach fits?
- AResponses API calls from their own servers
- BA prompt agent with function tools
- CA voice-based prompt agent with tools
- DA hosted agent running the team's code
Show the answer
Custom code and custom orchestration are the signature of a hosted agent: you bring the code and framework, and Foundry supplies the managed endpoint, scaling, and identity. Prompt agents fit production agents that do not need custom orchestration logic.
Next → 10 / 10 -
You’ve finished this set
That’s 10 questions on Foundry Agent Service at a Glance. In the app the ones you miss come back exactly when you’re about to forget them.
The whole course, on your phone
Lessons you can read, audio you can listen to on the way to work, and practice that remembers what you got wrong.
“This app is the ‘One Lake’ for preparing the certification exams. Really worth it.”