Home Certifications Editorials

Exam overview and study plan

Professional Agentic Architect PAA-Beta Practice Questions

Google Cloud Certified - Professional Agentic Architect (Beta)

Read questions, check verified answers and discuss the reasoning. Use reading options to shape your revision, or switch to the simulator for timed practice.

View 69 free questions

69 questions in this view

Page 1 of 12

Question 1

You are building a quality assurance agent in the Gemini Enterprise app. Your organization stores thousands of recorded customer support calls as audio files in Cloud Storage, alongside their transcripts and metadata. The agent must allow analysts to perform semantic searches across call conversations, filter results by structured call metadata, and return responses with direct links to the relevant audio recording segments. You need to configure the data store ingestion architecture while minimizing operational overhead. What should you do?

Ingest call transcripts with metadata as JSON Lines (JSONL) documents into an unstructured data store, and reference the Cloud Storage audio URIs in each entry.
Upload transcript embeddings into an Agent Retrieval vector index, and deploy a custom Cloud Run service to handle metadata filtering and audio retrieval.
Ingest raw audio files from Cloud Storage into an unstructured data store, and enable automated speech parsing to extract conversational embeddings.
Ingest call transcripts with metadata into BigQuery, and connect a Gemini Enterprise structured data store to execute parameterized queries.

Select an answer above, then submit your vote. Reading or practicing does not cast a vote.

Question 2

You are architecting a multitenant enterprise agentic app on Gemini Enterprise Agent Platform that queries internal BigQuery datasets and updates employee records across enterprise microservices. Security compliance requires tool invocations to execute under the authenticated identity of the calling employee to enforce data-layer access control list (ACL) policies and audit logging across downstream services. You want to implement the simplest solution. How should you implement this solution?

Configure Agent Gateway to sign JSON Web Tokens (JWT) using a shared service account to forward the service identity to downstream tools.
Configure Agent Gateway to use API keys for authentication to downstream microservices, with each key associated with a specific microservice to track its usage.
Implement a custom OAuth 2.0 authorization server that issues short-lived access tokens to Agent Gateway, and attach the tokens to tool invocation requests.
Configure Agent Gateway with OpenID Connect (OIDC) token exchange to propagate authenticated caller credentials to downstream service tools.

Select an answer above, then submit your vote. Reading or practicing does not cast a vote.

Question 3

You are using Antigravity CLI augmented with Agents CLI to manage the full development lifecycle of an enterprise agent built using Agent Development Kit (ADK). You have completed local prototyping, tool integration, and automated evaluation runs. You now need to govern the finalized agent across your organization so that security and administrative teams can discover, audit, and manage the agent. What should you do?

Invoke the google-agents-cli-deploy skill to execute the agents-cli deploy command.
Invoke the google-agents-cli-publish skill to execute the agents-cli publish command.
Invoke the google-agents-cli-observability skill to execute the agents-cli audit --bq-analytics command.
Invoke the google-agents-cli-scaffold skill to execute the agents-cli scaffold enhance --deployment-target agent_runtime command.

Select an answer above, then submit your vote. Reading or practicing does not cast a vote.

Question 4

You are configuring a developer workspace that uses Agents CLI alongside Antigravity CLI to manage the agent development lifecycle. You need to configure skill integration so that Antigravity CLI can discover and execute local ADK lifecycle skills at the workspace level. You also need to enforce a human-in-the-loop (HITL) policy in human mode, where human developers must manually inspect and approve production deployment commands before execution. What should you do?

Execute the agents-cli setup --workspace command in the project root directory to inject ADK development skills into the local workspace for detected coding agents.
Require developers to execute production deployment commands in human mode without noninteractive auto-approval flags.
Pass the --mode agent flag to all terminal invocations to force interactive prompts during CLI background tasks.
Modify the local .agents/skills/deploy/SKILL.md configuration file to include an autoApprove: false parameter under the human_mode metadata block to allow HITL.
Execute the agents-cli setup --global-only command, and set a AGENTS_CLI_AUTO_APPROVE=false parameter in the project .env file to grant coding agents direct production deployment rights with HITL as approver.

Select an answer above, then submit your vote. Reading or practicing does not cast a vote.

Question 5

You are building an analytics agent using Agent Development Kit (ADK) that will provide answers to questions based on data stored in BigQuery. You need to give the agent the ability to query data in BigQuery without adding integration code or servers that your team must maintain. You want authentication and access control to be handled by the platform. What should you do?

Generate an ADK OpenAPI toolset from the BigQuery discovery document.
Use the MCP Toolbox for Databases with a BigQuery source, and point the agent at its endpoint.
Publish a skill in Skill Registry to allow the agent to query BigQuery APIs programmatically.
Connect the agent to Google Cloud's remote MCP server for BigQuery.

Select an answer above, then submit your vote. Reading or practicing does not cast a vote.

Question 6

You are deploying a multimodal device-troubleshooting voice and chat agent built using Customer Experience Agent Studio (CX Agent Studio). Users upload a photo of their device to help diagnose hardware issues. The uploaded photo payload is sent to an external image-analysis endpoint through an OpenAPI tool call. In production, you observe that the agent frequently experiences timeout errors or pauses for up to ten seconds before responding, causing users to abandon sessions.

You need to resolve this tool invocation latency and timeout issue while maintaining a natural, high-quality user experience. What should you do?

Migrate the connection channel from the bidirectional BidiRunSession endpoint to REST-based polling, and isolate the image payload upload into a separate background thread.
Increase the timeout setting of the CX Agent Studio OpenAPI tool to at least ten seconds, and enable asynchronous tool calling in CX Agent Studio. Configure a transition instruction to immediately display a placeholder conversational response (e.g., "Let me take a look at that photo...") to keep the session alive while the background tool call executes.
Implement client-side image compression to reduce payload size, and configure a synchronous OpenAPI tool with a two-second timeout and client-side retry logic to process the request.
Implement a custom before_model_callback Python function in CX Agent Studio to intercept the user's photo upload. Write synchronous Python code in the callback to block the LLM request thread, perform the base64 image analysis, and write the diagnostic variables to the session context before releasing the model request.

Select an answer above, then submit your vote. Reading or practicing does not cast a vote.

Page 1 of 12

Build confidence for exam day with expert-reviewed questions and realistic practice. Your next certification starts here.

Company

About Contact Our Authors FAQ

Trust and policies

Methodology Editorial policy Terms and conditions Privacy policy

Contact

ExamPrepare LLC5203 Juan Tabo Blvd STE 2BAlbuquerque, NM 87111[email protected]

© 2026 ExamPrepare LLC. All rights reserved. All certification names, exam codes and logos are the property of their respective owners. All associated trademark and copyright rights belong to those owners. ExamPrepare is not affiliated with, endorsed by, or sponsored by any certification vendor.