UUS Practice Tests⌕ Search
IT Certifications · 1Z0-1157-26

Oracle Agentic AI Foundations Associate Exam Practice Test

The Oracle Agentic AI Foundations Associate Exam practice page focuses on building and evaluating AI agents that work through tasks, use tools, and interact with users. Its practice questions touch on output validation, model selection, arithmetic accuracy, workflow design, planning choices, tool-failure handling, and safeguards for consequential actions.

Exam overview

Exam providerOracle
Exam code1Z0-1157-26
CategoryIT Certifications

About the Oracle Agentic AI Foundations Associate Exam

Agent development requires more than producing plausible responses: teams must control how an agent interprets instructions, selects actions, and reports results. Practice scenarios explore structured validation when outputs must match allowed values, choosing a model suited to a routine classification task, and delegating exact calculations to reliable methods. They also address testing tool failures, checking returned data before making claims, and evaluating complete multi-step workflows rather than isolated replies. Other decisions involve selecting plans that account for action cost, making dependent stages observable for troubleshooting, obtaining evidence of user-perceived usefulness, and requesting confirmation before sending consequential communications. These concepts connect agent behavior to practical concerns such as correctness, traceability, and safe execution.

Sample Practice Questions

Review a selection from this exam's published US Practice Tests practice bank.

Question 1

An agent's prompt includes a list of valid category labels, but the model occasionally invents new labels. Which implementation most directly prevents invalid labels from being accepted by a downstream system?

  1. AValidate the output against an allowed-label set and handle invalid values
  2. BAsk the model to sound certain when choosing a label
  3. CIncrease the number of categories without updating the downstream system
  4. DRemove all instructions about the expected output
Show answerCorrect answer: A

Validation against the permitted set enforces the interface expected by downstream software. Prompt instructions can help, but validation catches outputs that do not meet the constraint.

Question 2

A team is choosing between a larger language model and a smaller one for a routine classification task. What is a sound way to make the choice?

  1. AEvaluate both on representative examples for quality, latency, and cost, then select one that meets requirements.
  2. BAlways choose the largest model because it is necessarily best for every task.
  3. CChoose whichever model has the most parameters, without testing the task.
  4. DUse the smaller model even if it fails required quality thresholds.
Show answerCorrect answer: A

Model selection is a task-specific tradeoff. Representative evaluation can reveal whether a less costly or faster model meets the needed quality, while avoiding unsupported assumptions based only on model size.

Question 3

A team is assessing an agent that invokes external tools. Which test type is especially useful for checking whether it handles tool failures appropriately?

  1. ASimulate timeouts, errors, and malformed tool responses, then inspect the agent's behavior.
  2. BTest only successful tool calls and assume failures will be handled the same way.
  3. CRemove all tool calls from testing, even though they are used in production.
  4. DMeasure only the length of the agent's final responses.
Show answerCorrect answer: A

Failure-path tests reveal whether the agent can recognize tool problems, avoid treating failed results as valid, and respond or recover appropriately. Success-only tests cannot establish that behavior.

Question 4

An agent is asked to add a set of expenses. What is the most reliable way to obtain the total?

  1. AAsk a language model to estimate the sum from the list
  2. BUse a deterministic calculation tool and verify the returned result
  3. CRound each expense to the nearest ten before adding
  4. DUse the number of expenses as an approximation of the total
Show answerCorrect answer: B

Arithmetic is best handled by a deterministic calculator or code tool rather than relying on a language model to generate a number. Checking the result also helps catch tool or input errors.

Question 5

An agent is being assessed on whether it can complete a workflow correctly across several steps. Which evaluation setup gives the clearest evidence of end-to-end task performance?

  1. AScore only whether the first response is grammatically correct
  2. BRun representative tasks from start to finish and assess whether the required outcomes were achieved
  3. CAsk the model whether it believes it performed well
  4. DMeasure only the number of tokens generated
Show answerCorrect answer: B

End-to-end evaluation checks the actual task outcome, including interactions among planning, tools, and intermediate steps. Fluency or self-assessment alone does not establish workflow success.

Question 6

An agent is given a task that can be completed in several possible ways, but some actions are costly. Which planning approach is generally preferable?

  1. AChoose a sequence of actions that is sufficient to meet the goal while avoiding unnecessary steps.
  2. BCall every available tool, whether or not its output is relevant.
  3. CRepeat the same action until the user stops the agent.
  4. DTake the most expensive action first because it may finish sooner.
Show answerCorrect answer: A

Goal-directed planning should account for whether actions are useful and proportionate to the task. Unnecessary tool calls can add cost, latency, and risk without improving the result.

Question 7

A tool returns a successful status code, but its output is an empty list. What should the agent do before telling the user that matching records were found?

  1. ATreat the success status as proof that records exist
  2. BInspect the returned data and report that no matching records were returned
  3. CInvent representative records to fill the empty result
  4. DRepeat the same request until the list is non-empty
Show answerCorrect answer: B

A successful tool call means the request was processed, not that it produced the desired data. The agent should interpret the actual result and avoid claiming findings that are not present.

How to use this practice test

The free practice session draws up to 20 random questions from the available 50-question bank. Answer the scenarios as you would assess an agent in operation: check output constraints, tool results, workflow stages, and action risks. Results appear after the session so you can identify which decisions need further review.

Preparation tips

  • For constrained outputs, distinguish prompt instructions from enforceable validation, and consider how a downstream system should reject values outside an allowed set.
  • Compare model choices against task needs such as classification accuracy, latency, and cost rather than assuming a larger model is always preferable.
  • Trace tool-use scenarios from invocation through returned data; distinguish a successful status from evidence that the requested records or result actually exist.
  • Practice mapping multi-stage workflows into observable steps, including failure handling, end-to-end evaluation, and checks before consequential actions such as sending an email.
Official sourceOfficial exam page ↗