Free Professional-Agentic-Architect – Google Cloud Certified - Professional Agentic Architect Practice Test Question

Select an option, then click Submit answer.

• ○

  Ingest the documents into an Agent Search data store with access controls (ACLs) synchronized from the source system, and connect the agent to that data store for grounded retrieval

• ○

  Fine-tune the Gemini model on the policy documents so the knowledge is built into the model

• ○

  Paste the most important policies into the agent's system instruction and rely on the model to respect confidentiality

• ○

  Store the documents in a Cloud Storage bucket and give the agent's service account read access to the entire bucket

Select an option, then click Submit answer.

• ○

  Add an instruction to the agent's system prompt telling it never to deploy to production

• ○

  Remove production IAM permissions from the developers who use the agent

• ○

  Configure the agent's tool access so only development-environment MCP servers and tools are exposed, excluding production deployment tools

• ○

  Review Cloud Audit Logs weekly to detect whether the agent performed a deployment

Select an option, then click Submit answer.

• ○

  A plugin that adds a security button to the IDE toolbar

• ○

  A subagent with a scoped security-review role and limited tool access that the main agent delegates review tasks to

• ○

  A rule stating that security issues must be treated as important

• ○

  A separate coding agent installation reserved for the security team

Select an option, then click Submit answer.

• ○

  Configure the agent in Gemini Enterprise Workflow Builder using only in-console settings

• ○

  Build the agent in CX Agent Studio with state-based pages and transition routes

• ○

  Call the Gemini API directly with hand-written HTTP requests and manage all session state manually

• ○

  Build the agent with the Agent Development Kit (ADK), defining tools as code and deploying the same agent to production

Select an option, then click Submit answer.

• ○

  The lowest per-token price, regardless of context size

• ○

  The smallest model that can be self-hosted on a single GPU

• ○

  Support for the most programming languages in code-generation benchmarks

• ○

  A large context window so the entire contract can be reasoned over in a single pass

Select an option, then click Submit answer.

• ○

  Grant the agent direct database write access and let the model compose raw SQL for the transaction

• ○

  Build a custom integration layer that implements the transaction with validated parameters and exposes it to the agent as a tool

• ○

  Split the transaction into separate manual steps performed by users outside the agent workflow

• ○

  Abandon the workflow because prebuilt tools cannot cover it

Select an option, then click Submit answer.

• ○

  A single Compute Engine VM that the team restarts manually after deployments

• ○

  Cloud Functions with conversation state kept in instance memory between calls

• ○

  A developer's workstation exposed through a tunnel for production traffic

• ○

  Agent Runtime, the managed Agent Platform runtime for deploying and scaling agents

Select an option, then click Submit answer.

• ○

  Increase the size of the golden dataset until the standard metrics become meaningful

• ○

  Score each draft with ROUGE against a small set of reference letters

• ○

  Rely on end-user thumbs-up and thumbs-down ratings as the compliance signal

• ○

  Build custom autoraters that score responses against the compliance rubric, calibrated on examples labeled by compliance experts

Select an option, then click Submit answer.

• ○

  Google Kubernetes Engine, so the team can operate scaling and session state itself

• ○

  Agent Runtime, which provides a managed runtime for agents with built-in scaling, sessions, and memory

• ○

  Cloud Run, with session and memory management reimplemented inside the application container

• ○

  A dedicated Compute Engine VM sized for peak load

Select an option, then click Submit answer.

• ○

  Replace the production version in place for all users at the same time

• ○

  Run the old and new versions permanently and let each user choose between them

• ○

  Deploy the new version and rely on user reports in the first week to surface problems

• ○

  Roll out the new version gradually with canary traffic and promote it only after it passes the evaluation gates against the golden dataset

Select an option, then click Submit answer.

• ○

  It automatically rewrites agent code to add tracking libraries at deployment time

• ○

  It scans source code repositories nightly and registers any agent-like code it finds

• ○

  It blocks all traffic that does not originate from a Cloud Run service

• ○

  As the mediation point for agent traffic, it observes and records which agents are active, letting the team track known agents and surface unregistered ones

Select an option, then click Submit answer.

• ○

  Keep the service account but add Model Armor screening as the only change

• ○

  Replace shared-account access with per-user identity propagation through Agent Gateway, so retrieval honors each employee's existing permissions, keeping registered tools governed in the Agent Registry

• ○

  Disconnect the agent from the corpus and let it answer from the model's training data instead

• ○

  Restrict the agent to senior staff only, who are trusted with the full corpus