Multi-Agent AI Solutions Expert (AI-500) - Full Practice Exam 1

1. A backend must call a downstream API within a signed-in user's delegated authorization rather than with unrestricted app-only authority. Which identity flow should be evaluated where supported?
2. A business rule forbids refunds above a threshold without approval. Where should that rule be enforced?
3. A candidate release passes prompt-only tests but changes tool contracts and authorization middleware. Which further validation is needed?
4. A durable workflow resumes from a checkpoint. How should it handle a previously completed external action?
5. A guardrail is applied only to final text, but tools can mutate records during execution. What is missing?
6. A human approves an agent's proposed purchase, but the agent then changes the amount and recipient. What should happen?
7. A judge model scores outputs highly, but expert review finds a systematic error. What should the evaluation process do?
8. A long-running workflow must resume after a process restart without forgetting completed steps. Which framework capability should be evaluated?
9. A new orchestration version should receive limited traffic while the previous version remains available. Which release strategy fits?
10. A payment tool times out after the backend may have committed. Which integration property is required for safe retries?
11. A peer agent advertises a capability through a supported A2A interface. What should the coordinator validate before delegating sensitive work?
12. A repeatedly summarized conversation gradually changes a customer's approved delivery address. What should be investigated?
13. A replay must reproduce a previous failed workflow for authorized debugging. Which records are important?
14. A retrieved page instructs an agent to upload confidential data to an outside endpoint. What should the security architecture enforce?
15. A semantic cache is shared by agents serving several tenants. Which values must influence safe reuse besides semantic similarity?
16. A specialist agent receives source documents and instructions from an orchestrator. How should the prompt distinguish them?
17. A system needs conversation-local state, shared task state, and durable semantic memory. Which architecture is appropriate?
18. A task is a single bounded transformation that one evaluated model call performs reliably. What should an expert consider before adding multiple agents?
19. A team performs adversarial security tests of an agent with real production mutation tools. What should the test design do?
20. A team proposes caching the success result of a mutating tool and returning it for semantically similar future requests. What is the main concern?
21. A test set contains only cooperative users and correct tool responses. Which additions best improve coverage?
22. A tool returns a syntactically valid result with an impossible negative shipment weight. What should the workflow do?
23. A workflow crosses multiple agents and services. Which observability design supports diagnosing one user's failed operation?
24. A workflow must first retrieve an order, then evaluate its eligibility, and only then prepare an authorized action. Each step depends on the previous result. Which orchestration pattern is the best starting point?
25. A workflow performs several external business operations and fails halfway through. Which architecture should be evaluated where the business supports it?
26. A workflow reaches its configured execution budget before finishing. What should it do?
27. A workflow version changes its persisted state schema. What must a rollback strategy consider?
28. A workflow's cost increases despite a stable number of user requests. Which accounting is most useful?
29. Adding more parallel agents increases throttling and total latency. Which optimization should be evaluated?
30. Agent A may read customer records, while Agent B may submit approved refunds. Which identity design best limits lateral impact?
31. Agents run in a private network but call a knowledge service. Which two boundaries must remain enforced? Select TWO.
32. An agent forgets an early nonnegotiable constraint after many turns. Which mitigation is most appropriate?
33. An agent release passes evaluations but production deployment requires an operations approval. What should the pipeline do?
34. An agent returns a relevant and fluent answer that invents an unsupported policy exception. Which evaluation is most directly needed?
35. An agent's response format changes between releases and breaks its coordinator. Which practice should be implemented?
36. An architecture integrates tools and independent peer agents. Which two protocol assignments should be evaluated? Select TWO.
37. An MCP server exposes a tool for updating records. Which boundary must still be enforced?
38. An orchestration graph cycles between two agents without task progress. Which safeguards should be added?
39. Conversation history exceeds the model's context budget. What should compaction preserve explicitly?
40. Different subtasks require exact calculation, document retrieval, and language synthesis. Which component-selection strategy is appropriate?
41. Every specialist passes its unit tests, but the full workflow fails after a handoff. What additional evaluation is required?
42. Independent agents repeatedly call a shared rate-limited API. What should orchestration implement?
43. Long-term memory contains a user's obsolete preference and a newer correction. Which design is appropriate?
44. Multi-agent infrastructure must be promoted consistently through development, test, acceptance, and production. Which practice best supports control?
45. Parallel tasks perform external writes. One task fails while others may still run. What must the orchestrator handle explicitly?
46. Several agents consume one enterprise knowledge corpus, but they serve users with different permissions. What must retrieval enforce?
47. Several agents need consistent logging, authorization, and exception handling. What should the implementation evaluate?
48. Several specialists can propose changes to the same business record. The system needs one accountable component to validate proposals and authorize execution. Which architecture fits?
49. Some agent tools still need external credentials. Which secret strategy is appropriate?
50. Three independent research tasks can run concurrently before a coordinator combines their results. Which design is appropriate?

Comments — 0