Can you design an agentic system that is useful, observable, secure, and affordable—not just impressive in a demo?

Prepare for the Google Cloud Professional Agentic Architect certification with 600 original single-answer questions across six tests of 100 questions each. This is a professional, technical exam: it expects real design judgment about low-code agents, coding agents, custom agents built with ADK, retrieval, evaluation, deployment, security, and governance.

WHAT YOU GET

• Six original 100-question tests

• Four alternatives and ONE best answer for every item

• An explanation for every option plus a case-specific overall rationale

• Guide-section anchors and local study-map skill mappings

• Five focused learning tests followed by one integrated review

• Offline review copies alongside instructor CSV files

THE SIX TESTS

1. LOW-CODE AGENTS AND CODING AGENTS

Configure pages, transition routes, and event handlers; write system instructions and prompt templates; connect enterprise data and multimodal sources; and control release changes. Then configure coding agents with MCP servers, skills, plugins, hooks, rules, and subagents; use isolated sandboxes; and plan safe refactors and vulnerability fixes.

2. CUSTOM AGENTS: MODELS, ADK, MEMORY, RAG, AND RETRIEVAL

Choose models by capability, cost, latency, control, and security. Build with the Agent Development Kit, define tools and callbacks, manage sessions, state, memory, and artifacts, and design retrieval with chunking, embeddings, reranking, grounding, and managed vector services.

3. CUSTOM AGENTS: IDENTITY, ORCHESTRATION, AND REGISTRY

Apply agent identity and least privilege, principal access boundaries, registry and MCP capability integration, Agent Cards and A2A interoperability, sequential, parallel, graph, routing, and handoff patterns, and runtime coordination.

4. EVALUATING AND DEPLOYING AGENTIC WORKFLOWS

Build golden datasets and edge cases, evaluate responses, retrieval, and tool trajectories, run continuous evaluation and regression, choose Agent Runtime, Cloud Run, or GKE, and troubleshoot drift, loops, tool latency, and cost.

5. SECURING AND GOVERNING AGENTIC WORKFLOWS

Secure tool execution with scoped OAuth tokens, constrain identities with principal access boundaries, observe and control traffic through Agent Gateway, enforce policy with Agent Registry and Model Armor, and design human-in-the-loop guardrails and data protections.

6. INTEGRATED ARCHITECT SCENARIOS AND FINAL REVIEW

Apply all five published sections in an expanded mixed review: 13 low-code, 17 coding agents, 33 custom agents, 22 evaluation and deployment, and 15 security and governance questions.

COVERAGE THAT IS EXPLICIT ABOUT ITS LIMITS

All five published sections and 32 local study-map skills have mapped practice. Numeric skill IDs are local study aids, not official guide numbering. Mapping does not mean every consideration has equal depth or that questions were independently calibrated.

Tests 1–5 intentionally emphasize particular areas, especially Developing Custom Agents (33% of the exam and 200 questions here). Neither those tests nor the aggregate 600-question bank follow official weights. Test 6 approximates the published percentages with a 100-question integer allocation. It is an expanded learning review, not an exact exam replica.

HOW THE CERTIFICATION ACTUALLY WORKS

The certification is hybrid and has two required parts. Part 1 is a proctored multiple-choice exam delivered by Pearson. The beta exam was 3 hours with roughly 80 questions and cost $120; Google says general availability opens in mid-November 2026 with a 2-hour session at $200 list price. Part 2 is a separate set of hands-on labs in Google Skills, available only after you pass the multiple-choice exam. Passing the questions alone does not earn the credential.

The credential is valid for 1 year and has no prerequisites. Google recommends 3+ years of hands-on experience building, testing, deploying, and managing cloud solutions, including 1+ year building agentic solutions on Google Cloud.

HOW PRACTICE DIFFERS

Every question here is single-answer. This package does not simulate the official item count, Pearson delivery, adaptive behavior, or the required hands-on labs. Suggested practice duration is 180 minutes per 100-question test; the optional 80% study target is custom learning guidance, not an official pass mark. Google does not publish a raw percentage pass threshold for this exam, and no score conversion is implied.

Concepts and option definitions recur for reinforcement. Six hundred distinct question texts do not mean six hundred independent concepts or six independent readiness estimates. No score, calibrated difficulty, lab completion, or certification result is guaranteed.

VERSION BOUNDARY

Baseline: the current official English Professional Agentic Architect exam guide (five sections; 13/17/33/22/15). The beta multiple-choice window ran September 8–30, 2026 and is closed. Google says general availability opens in mid-November 2026, with a 2-hour proctored session and the same two-part structure. The GA exam guide and format may change; this package makes no claim about an unpublished future blueprint.

Product capabilities, quotas, regional availability, and interfaces change. Questions therefore test architectural reasoning and ask you to verify current supported behavior instead of memorizing a fixed console path or price.