Google Cloud Professional Agentic Architect - Full Practice Exam 1 - Beta Blueprint

1. A business team wants a low-code enterprise agent using supported business connectors and configurable instructions. Which approach best fits?
2. A coding agent generated a command containing untrusted user input. The command will execute in a shell. Which approach best fits?
3. A coding agent may propose a production deployment, but the workflow requires human authorization for publishing. Which approach best fits?
4. A coding agent must patch a vulnerability without altering unrelated behavior. Which approach best fits?
5. A coding agent needs a supported standardized connection to a set of external development tools. Which approach best fits?
6. A coding agent performs a large refactor and changes generated files unrelated to the requested feature. Which approach best fits?
7. A coding agent should format and test changes consistently before they are accepted. Which approach best fits?
8. A coding assistant should inspect a repository and propose changes, but its execution must not access unrelated host files or production credentials. Which approach best fits?
9. A company introduces an external MCP server that can access internal resources. Which approach best fits?
10. A deployed agent must be distinguishable from other workloads when it calls Google Cloud resources. Which approach best fits?
11. A deployment should test a new agent version with limited exposure and preserve a known-good fallback. Which approach best fits?
12. A developer wants coding-agent suggestions to use project-specific patterns and conventions. Which approach best fits?
13. A development workflow needs a managed environment with defined access and resource controls for coding tasks. Which approach best fits?
14. A large manual needs retrieval at passage level while keeping enough context to interpret each result. Which approach best fits?
15. A long-term agent memory stores personal information beyond the approved retention period. Which approach best fits?
16. A low-code agent needs current employee-specific data but different employees have different source permissions. Which approach best fits?
17. A low-code agent should acknowledge that no approved document supports an answer. Which approach best fits?
18. A low-code assistant must answer from a governed enterprise document collection. Which approach best fits?
19. A low-code assistant should process images and text from enterprise content. Which approach best fits?
20. A low-code customer-service agent must follow explicit conversational states, collect required fields, and transition according to defined events. Which approach best fits?
21. A managed model endpoint meets the task and data requirements, while the team lacks model-serving expertise. Which approach best fits?
22. A multi-user agent caches retrieved documents without including user access context in the cache key. Which approach best fits?
23. A new tool schema is incompatible with an existing deployed agent version. Which approach best fits?
24. A platform needs discovery and governance metadata for deployed agents and approved capabilities. Which approach best fits?
25. A production agent needs a managed runtime appropriate to supported agent deployment while minimizing custom orchestration infrastructure. Which approach best fits?
26. A RAG assistant's answer quality drops after a corpus update. The team must separate retrieval failure from generation failure. Which approach best fits?
27. A retrieval pipeline returns many semantically related passages, but the best evidence ranks too low for the final prompt. Which approach best fits?
28. A retrieved web page tries to instruct an agent to transmit confidential context to an external address. Which approach best fits?
29. A reusable agent extension needs access to one issue tracker project, not the entire organization's administration API. Which approach best fits?
30. A security design assumes that adding safety filters eliminates the need for IAM on agent tools. Which approach best fits?
31. A sensitive task requires a self-hosted model in an approved environment, and the team accepts operational responsibility. Which approach best fits?
32. A sensitive tool action needs explicit user approval after the proposed operation is fully described. Which approach best fits?
33. A simple workflow fits native supported connectors, but a proposal adds a custom server solely to duplicate them. Which approach best fits?
34. A source document is deleted after its retention period, but retrieval still returns an old indexed copy. Which approach best fits?
35. A team adds a new coding-agent plugin from an unknown source that can run arbitrary local commands. Which approach best fits?
36. A team needs repeatable expected examples for regression testing across agent changes. Which approach best fits?
37. A team repeatedly asks a coding agent to follow the same repository-specific procedure. Which approach best fits?
38. A tool invocation succeeds technically but updates the wrong customer record. Which approach best fits?
39. A tool performs an irreversible operation and must not run simply because generated text resembles a request. Which approach best fits?
40. A tool receives a customer identifier from model output and accesses a tenant database. Which approach best fits?
41. A tool returns an instruction to reveal another user's private session history. Which approach best fits?
42. A tool service becomes unavailable and model retries create growing request pressure. Which approach best fits?
43. A workflow combines cheap classification with more expensive reasoning only for complex requests. Which approach best fits?
44. A workflow has branching state transitions that depend on prior validated results. Which approach best fits?
45. A workflow must execute validation before a consequential action and may not reorder the steps. Which approach best fits?
46. An agent can read billing data but must never modify payment settings. Which approach best fits?
47. An agent claims a patch is correct because its own explanation sounds convincing, but no checks ran. Which approach best fits?
48. An agent evaluation checks only the final text, but the agent sometimes invokes the wrong external operation. Which approach best fits?
49. An agent has access to many tools but only two are needed for the current approved task. Which approach best fits?
50. An agent is packaged as a stateless HTTP container and does not require Kubernetes-specific controls. Which approach best fits?
51. An agent is slow, and the team must distinguish model latency from retrieval and external tool delays. Which approach best fits?
52. An agent logs full prompts containing secrets and confidential records to a broadly accessible destination. Which approach best fits?
53. An agent must screen supported prompts and responses for defined harmful or sensitive content risks. Which approach best fits?
54. An agent needs a standard interface to invoke authorized external tools and resources. Which approach best fits?
55. An agent needs freshness guarantees for a rapidly changing inventory API rather than a nightly indexed document snapshot. Which approach best fits?
56. An agent needs temporary conversational context for one user interaction sequence. Which approach best fits?
57. An agent receives a repository document that asks it to upload credentials to an external address. Which approach best fits?
58. An agent repeatedly calls the same tool without making progress. Which approach best fits?
59. An agent retries a tool call after a network timeout, but the first call may already have created the ticket. Which approach best fits?
60. An agent should cite the exact source supporting its conclusion rather than a merely related document. Which approach best fits?
61. An agent should only use approved model versions and tools after evaluation. Which approach best fits?
62. An agent should repeat an improvement step only until a measurable criterion or iteration limit is reached. Which approach best fits?
63. An agent should route a conversation to a human when a required condition or safety threshold is reached. Which approach best fits?
64. An agent should use a small model for a straightforward task when it meets approved quality and safety criteria. Which approach best fits?
65. An agent update changes prompts, tools, and model version simultaneously, making regressions hard to diagnose. Which approach best fits?
66. An agent workflow must recover after a process restart without losing which external actions already completed. Which approach best fits?
67. An agent workload requires Kubernetes-specific policy, sidecars, and custom runtime orchestration. Which approach best fits?
68. An agent's cost increases after a prompt change despite stable user traffic. Which approach best fits?
69. An agent's document embeddings and query embeddings come from incompatible model versions. Which approach best fits?
70. An agent's evaluation dataset omits prompt injection and unavailable-tool scenarios. Which approach best fits?
71. An agent's prompts and evaluations should be reproducible during an incident review. Which approach best fits?
72. An application needs code-level orchestration of agent tools, state, and execution with a supported open framework. Which approach best fits?
73. An assistant should retain approved useful information across separate sessions with explicit privacy and retention controls. Which approach best fits?
74. An author changes a low-code agent prompt and wants evidence that important existing tasks still work. Which approach best fits?
75. An automatic judge rates responses, but its scores disagree with expert reviewers on consequential cases. Which approach best fits?
76. An enterprise needs governance over which tools agents can reach and how approved calls are controlled. Which approach best fits?
77. An existing low-code agent gives verbose answers and omits required steps, while its source documents are correct. Which approach best fits?
78. Several agents share one powerful credential, preventing reliable attribution of individual actions. Which approach best fits?
79. Several independent retrieval tasks can run simultaneously before their results are combined. Which approach best fits?
80. Two independently implemented agents need a supported protocol for task-oriented communication and collaboration. Which approach best fits?

Comments — 0