Google Cloud Professional Agentic Architect

Objective: Design, build and govern a production multi-agent platform that combines a low-code customer front door with ADK agents, grounded answers and end-to-end security controls.

What you will build

• Build a low-code front door in CX Agent Studio, with a flow-based agent for order status and returns that uses pages, transition routes and event handlers, guided by few-shot and chain-of-thought templates.
• Ground answers in enterprise knowledge by ingesting policy PDFs, product images and support-call audio into Agent Search and a RAG Engine corpus, with reranking for product lookup.
• Develop an ADK coordinator agent that delegates to billing, returns and inventory sub-agents using sequential and parallel workflows, with the inventory agent exposed over A2A.
• Connect tools through MCP: a Google Cloud MCP server for BigQuery and a custom MCP server for a third-party CRM deployed on Cloud Run and registered in Agent Registry.
• Add sessions and long-term memory with managed sessions and Memory Bank on Agent Runtime.
• Secure and govern the system with Agent Identity, OAuth 2.0 through Auth Manager, a PAB policy, Agent Gateway egress policies, Model Armor templates and HITL approval for refunds above a threshold.
• Evaluate and deploy with an ADK evalset of 50+ golden and edge cases, a CI evaluation gate, deployment to Agent Runtime, and Cloud Logging and Cloud Trace dashboards.

Deliverables

• Architecture and ADR — a system diagram and decision record covering model, runtime and security choices
• Working agent system — CX Agent Studio front door plus ADK coordinator and sub-agents deployed on Agent Runtime
• Evaluation report — evalset results, the CI evaluation gate and monitoring dashboards
• Security and governance pack — Agent Identity, PAB, Agent Gateway policies, the Model Armor template and HITL evidence

Objective: Use customized coding agents in secure sandboxes to modernize a legacy application, then build, deploy and continuously evaluate an internal ADK agent with Agents CLI.

What you will build

• Set up secure coding-agent sandboxes on Cloud Workstations with restricted egress, plus Antigravity's terminal sandbox and approval rules for risky commands.
• Customize Antigravity with team rules, reusable skills, a lifecycle hook that runs lint and security scans, a security-reviewer subagent, and a plugin that bundles them for the team.
• Modernize a legacy application with Antigravity and Claude Code on Google Cloud: add tests, refactor a module into a service, reduce container size and cold starts on Cloud Run, and patch an injection flaw and a vulnerable dependency.
• Build a "DevOps assistant" ADK agent with Agents CLI that answers deployment questions from Cloud Logging and BigQuery data, then deploy and evaluate it through Agents CLI.
• Pick and justify the runtime by comparing Cloud Run, GKE and Agent Runtime on cost, latency and operational effort.
• Create a continuous evaluation pipeline that scores the assistant's tool execution against success criteria and blocks releases on regression.

Deliverables

• Source repository — agent code, Antigravity rules, skills, hooks, subagent and plugin definitions, and MCP configuration
• Modernization report — before/after test coverage, vulnerabilities fixed, image size and latency
• CI/CD pipeline — workflow configuration with automated evaluation gates and stored results
• Architecture Decision Record — 600–900 words on model selection, runtime choice, security controls and trade-offs

Evaluation Rubric (aligned to exam weightage)

1. Building agents using low-code tools (13%) — workflow design, prompt templates and grounded data access.
2. Using coding agents (17%) — sandbox security, customizations, refactoring and patch quality.
3. Developing custom agents (33%) — ADK design, memory, RAG quality, MCP/A2A and multi-agent coordination.
4. Evaluating and deploying (22%) — golden dataset coverage, CI evaluation, runtime choice and observability.
5. Securing and governing (15%) — Agent Identity, PAB, Agent Gateway, Model Armor and HITL.