Google Cloud Professional Agentic Architect practice questions, page 3 of 3, with answer review and explanations.
Question 11
Your organization maintains an internal service registry that lists the endpoints exposed by various teams across the organization. Endpoints are added and removed throughout the day. Your team uses Antigravity to write client code against the endpoints. You discover that the agent is generating code referencing stale, decommissioned endpoints. You need Antigravity to dynamically verify active endpoints from the registry in real time while generating code. What should you do?
Create an MCP server, and connect Antigravity to the registry using the MCP.
Define a custom Antigravity subagent whose role is endpoint lookup.
Add the endpoint list to an Antigravity workspace rule.
Add an Antigravity PostToolUse hook that checks generated calls against the registry.
Question 12
You are deploying an agent on Gemini Enterprise Agent Platform in a dedicated runtime Google Cloud project, Project-A. The agent needs to retrieve database API credentials from Secret Manager in a centralized governance project, Project-B, and query an Agent Search data store also hosted in Project-B. You are using Agent Identity to govern the agent’s permissions across projects. You need to grant the minimum required cross-project IAM access while adhering to the principle of least privilege. What should you do? (Choose two.)
Grant the roles/secretmanager.user IAM role to the agent’s identity at the Project-B project level.
Grant the roles/secretmanager.secretAccessor IAM role to the agent’s identity scoped to the database resource in Project-B.
Grant the roles/discoveryengine.editor IAM role to Project-A’s agent identity at the Project-B project level.
Grant the roles/discoveryengine.viewer IAM role to the agent’s identity scoped to the target Agent Search data store resource in Project-B.
Grant the roles/iam.serviceAccountTokenCreator IAM role to Project-B’s agent identity to allow cross-project impersonation.
Question 13
You are deploying a custom AI assistant written in Python using Agent Development Kit (ADK). You want to maximize developer velocity by deploying the agent code with minimal infrastructure plumbing, while ensuring the agent has built-in, out-of-the-box access to manage conversational session history and memory persistence. How should you deploy this agent?
Deploy the agent to Agent Runtime with Agent Platform Sessions and Agent Platform Memory Bank.
Deploy the agent as a containerized workload on Cloud Run. Configure an external BigQuery table to log and retrieve session messages.
Package the agent logic into a standard container. Deploy the container to Google Kubernetes Engine (GKE) with Agent Platform Sessions and Agent Platform Memory Bank.
Deploy the agent on Cloud Run. Attach a Bigtable instance to manage conversational state and session history.
Question 14
You are using CodeMender on Gemini Enterprise Agent Platform to remediate a critical SQL injection vulnerability in an enterprise Python web app. You need the coding agent to generate a production-ready patch that resolves the underlying vulnerability while keeping the existing app functionality intact. What should you do?
Configure CodeMender’s patch generator to execute dynamic database schema migrations to sanitize incoming string inputs before query execution.
Generate a patch using CodeMender’s automated remediation workflow, and use its built-in LLM-as-a-judge framework to validate program equivalence against the generated code diff.
Grant CodeMender’s execution identity the roles/iam.securityUser IAM role to run commands.
Instruct CodeMender to wrap dynamic SQL queries in try-except error handling blocks and retry failed queries using the default fallback parameters.
Question 15
You are architecting a continuous evaluation pipeline for an autonomous agent deployed on Agent Runtime. The agent uses multiple valid, non-linear sequences of function calls to resolve user requests instead of using hardcoded golden trajectories. You want to continuously assess the semantic validity of tool selection and parameter generation without using hardcoded reference files. What should you do?
Configure the pipeline to use custom function metrics in the gen AI evaluation service.
Configure the pipeline to use the managed types.RubricMetric.TOOL_USE_QUALITY metric in the gen AI evaluation service.
Configure the pipeline to use the managed types.RubricMetric.FINAL_RESPONSE_QUALITY metric in the gen AI evaluation service.
Export the agent’s intermediate tool calls to BigQuery, generate vector embeddings using the ML.GENERATE_EMBEDDING function, and execute the VECTOR_SEARCH command.