AI-901 - Implement AI solutions by using Microsoft Foundry (58% of the exam) - Section 2.1
Implement generative AI apps and agents by using Foundry, including prompts, deploying and interacting with a model, the Foundry SDK, and single-agent solutions.
Writing effective system and user prompts for generative models; deploying a model and interacting with it in the Foundry portal; building a lightweight chat client with the Foundry SDK; and creating and testing a single-agent solution in the Foundry portal. The reliable single-best distractions are choosing the portal where the SDK is required (or the reverse), and collapsing an agent (which plans and calls tools or knowledge to complete a task) into a plain generative chat completion.
system and user promptsMicrosoft Foundry portalFoundry SDKdeploy a model in Foundrysingle-agent solutionagent client application
Practice question for this objective
lock_openFree sampleImplement AI solutions by using Microsoft Foundryhard
A developer is deciding between building a single-agent solution and simply calling a deployed model with a plain chat completion request. Which statement correctly describes when the single-agent solution is warranted?
AA single agent is warranted when the task requires planning steps and calling tools or knowledge to act, whereas a plain chat completion only returns a generated reply.check_circle Correct
BA single agent is warranted whenever the reply needs to be longer than a plain chat completion is able to return in one response.
CA plain chat completion is warranted whenever the answer must be in British English, because an agent cannot be told which language to use.
DThe two are interchangeable, and the only real difference is which button is selected in the Foundry portal when the model is deployed.
Distinguish a single-agent solution, which plans and calls tools to act, from a plain chat completion that only returns a reply. A plain chat completion returns generated text from the model, but a single-agent solution can plan a sequence of steps and invoke tools or knowledge to carry out actions towards a goal, which is the capability that justifies building an agent.
Why A is correct: Correct: an agent adds planning and the ability to call tools or knowledge sources to complete a goal, while a plain chat completion just generates a reply from the model.
Why B is wrong: Response length is not what separates the two; both approaches ultimately produce model output, so length alone is not a reason to build an agent.
Why C is wrong: This is a distraction: language and tone are set through instructions in either approach, so an agent is perfectly able to reply in British English.
Why D is wrong: It is tempting to treat them as the same feature with different labels, but they differ in capability: an agent can plan and act, which a plain completion cannot.
Exam traps in Implement AI solutions by using Microsoft Foundry
Answers that look right on this material and are not. Each one is a distractor from a different question in the AI-901 bank for this domain.
The first uses the Foundry SDK and the second uses the Foundry portal.
errorWhy it is wrong: Tempting because both names are valid but reversed; interactive browser work is the portal, and code-based calls are the SDK.
Open the Foundry portal playground each time and paste tickets into the agent by hand for the service to read.
errorWhy it is wrong: The playground lets a person interact with the agent and is tempting because it is where the agent was tested, but manual pasting cannot serve an automated back-end workflow.
The Foundry SDK, because it exposes catalogue browsing and deployment through client library methods.
errorWhy it is wrong: The SDK is tempting because it also reaches the catalogue and deployments, but it is a code-based surface; the stem specifies working without writing code.