Google Cloud has entered a new certification territory, and this one is not simply about knowing cloud services. The new Professional Agentic Architect certification is built around a technology shift that is changing how AI applications are designed, deployed, and managed: AI agents that can reason, use tools, interact with data, and work together to complete complex tasks.
But there is something particularly interesting about this new exam. It isn’t testing only whether you understand agentic AI concepts. Google Cloud says the certification assesses capabilities ranging from building agents with low-code tools and coding agents to developing custom agents, evaluating and deploying agentic workflows, and securing and governing those workflows. The assessment itself combines a proctored multiple-choice exam with hands-on labs, making the practical side of the certification just as important as knowing the theory.
And here is where things get interesting: what exactly does Google Cloud expect an Agentic Architect to know, what makes this exam different from conventional cloud certifications, and which technologies and skills sit behind the certification? If you’re considering this new credential, there are several details you will want to understand before you start preparing. Let’s uncover what the Professional Agentic Architect exam is really about, who it is designed for, what you can expect from the assessment, and what it takes to prepare for this emerging certification.
What Is the NEW Google Cloud Professional Agentic Architect Exam?
Google Cloud has introduced a professional-level certification built around a different kind of architecture problem: designing, building, evaluating, deploying, securing, and governing agentic workflows. This is not simply a generative AI theory exam. The current beta combines a proctored multiple-choice assessment with hands-on labs.
Why Google Cloud created an Agentic Architect certification
The move from chatbots to agents changes the architecture question. A conventional application usually follows a largely deterministic path. An agentic system can reason over a task, select tools, retrieve information, hand work to another agent, maintain state, and continue operating across multiple steps.
Google Cloud describes the Professional Agentic Architect as a technical practitioner who designs and manages autonomous, AI-driven agentic workflows in Google Cloud. The role combines LLM knowledge, agent design patterns, coding, data integration, and cloud architecture decisions around reliability, performance, cost, security, and scalability.
The important shift
The credential is not centered on knowing one model or one prompt framework. It is centered on making agentic systems work as real cloud systems: connecting data, choosing models, coordinating agents, evaluating behavior, deploying at scale, and putting security and governance around the system.
Think of the credential as the bridge between “I can build an AI agent” and “I can architect an agentic system that an enterprise can actually operate.”
Exam vs. certification: this is a two-part model
This is the detail most quick summaries miss. Google Cloud says the Professional Agentic Architect certification will consist of two parts: a proctored multiple-choice exam delivered through Pearson and hands-on labs delivered through Google Skills. The multiple-choice exam assesses conceptual knowledge, system design choices and architectural standards. The labs validate hands-on execution and coding ability.
Part 1 · Proctored exam
- About 80 multiple-choice questions in the current beta
- 3-hour duration
- Online-proctored or onsite-proctored delivery
- Concepts, system design choices and architecture standards
- English language
Part 2 · Hands-on labs
- Delivered through Google Skills
- Validates hands-on execution and coding ability
- Candidates who pass the multiple-choice beta are automatically notified when labs are available
- Practical execution therefore matters to the certification path
Do not prepare as though this were only a theory exam
The lab component exists specifically to validate hands-on work. Practical experimentation, coding, deployment and troubleshooting belong in your learning plan.
| Current beta detail | Google Cloud’s published information |
|---|---|
| Duration | 3 hours |
| Multiple-choice format | ~80 questions |
| Language | English |
| Delivery | Online-proctored or onsite-proctored |
| Prerequisites | None |
| Recommended experience | 3+ years cloud solutions + 1+ year agentic solutions using Google Cloud |
| Beta fee | $120 + applicable tax |
| Retail price stated by Google | $200 USD |
The five exam domains and their weightings
The official exam guide organizes the assessment into five domains. The biggest concentration is Developing custom agents at approximately 33%, followed by evaluating and deploying workflows at 22% and coding agents at 17%.
One number to remember: ~33%
Developing custom agents is the largest single domain. It brings together model selection, ADK, memory and sessions, RAG and vector retrieval, agent permissions, integrations, and multi-agent orchestration.
What does each domain actually cover?
Instead of treating the exam guide as a long list of technologies, read it as an architecture lifecycle. Use the tabs below to see the role of each domain.
Section 1 · Building agents using low-code tools
Configure behavior and connect enterprise information without writing every component from scratch.
- State-based workflows, pages, transition routes and event handlers with Gemini Enterprise tools.
- System instructions and in-console prompt templates.
- Secure connection to enterprise data through tools such as Agent Search.
- Ingest and process unstructured multimodal data such as video, audio and images.
Section 2 · Using coding agents for application development
Agents that help engineers build, refactor and secure software.
- Configure coding agents with MCP servers, custom skills and tools.
- Use secure sandboxes such as GKE, Cloud Workstations and Antigravity.
- Refactor source code, optimize runtimes and patch application-layer vulnerabilities.
- Create skills, plugins, hooks, rules and subagents for enterprise workflows.
- Augment workflows with Agents CLI.
Section 3 · Developing custom agents
The largest domain: architecture, coding, data retrieval and multi-agent coordination.
- Select models and hosting approaches with cost, security and architecture in mind.
- Build custom agents with open-source libraries such as ADK.
- Configure sessions and memory with Agent Platform Memory Bank and managed sessions.
- Design RAG pipelines, embeddings, similarity scoring and reranking.
- Configure agent permissions with Agent Identity.
- Use Agent Registry and Google Cloud MCP Servers for prebuilt and custom capabilities.
- Coordinate multi-agent workflows using MCP and A2A, with sequential, parallel and graph workflows.
Section 4 · Evaluating and deploying agentic workflows
A production agent has to be measurable, observable, reliable and economically sustainable.
- Create evaluation test sets with golden data, prompts and edge cases.
- Build continuous evaluation pipelines against success criteria.
- Use ADK evaluation tooling, Gen AI evaluation services and custom autoraters.
- Evaluate response and retrieval quality against golden datasets.
- Select runtimes such as Agent Runtime, Cloud Run and GKE based on requirements and cost.
- Troubleshoot drift, tool latency, reasoning loops and system failures.
- Monitor and optimize performance, reliability and cost.
Section 5 · Securing and governing agentic workflows
The agent becomes an enterprise actor. Identity, boundaries, gateways and safety become architecture concerns.
- Secure agent-to-tool calls with authentication and OAuth 2.0.
- Configure principal access boundary policies with Agent Identity.
- Use Agent Gateway to monitor traffic and track agents.
- Apply policy enforcement with Agent Registry and Model Armor.
- Design guardrails and human-in-the-loop controls.
- Secure data access and identity propagation.
The agentic architecture lifecycle
The exam blueprint becomes easier to understand when you follow an agent from design to production: configure, build, connect, orchestrate, evaluate, deploy, monitor, secure and govern.
The question is not simply “Which AI product does this?” It is “What architecture makes sense for this workload, requirement and risk?”
A scenario to think about
Imagine an enterprise agent that retrieves financial policy data, calls an internal API, hands part of the work to another specialized agent, and must keep every tool invocation inside a controlled security boundary. The architecture now has retrieval, identity, orchestration, evaluation, observability and policy concerns at the same time.
Which Google Cloud tools and concepts are in scope?
Group the official tool list by the job it performs in an agentic system. That creates a more useful mental model than memorizing product names independently.
| Architecture concern | Examples in the scope | Question to ask |
|---|---|---|
| Build | ADK, Antigravity, Agents CLI | How do I create and customize the agent? |
| Knowledge | Agent Search, RAG Engine, Agent Retrieval, Vector Search | How does the agent find authoritative information? |
| Identity | Agent Identity, Agent Registry, Auth Manager | What may the agent access and on whose behalf? |
| Orchestration | MCP, A2A, Agent Runtime, policies | How do agents coordinate and use tools? |
| Runtime | Agent Runtime, Cloud Run, GKE | Where should the workload run? |
| Evaluation | Eval sets, golden data, autoraters | How will I know the agent works? |
| Safety | Agent Gateway, Model Armor, HITL, PAB | What happens when an action is risky? |
| Observability | Cloud Logging, Cloud Trace | How do I diagnose drift, latency or failure? |
Who is the Professional Agentic Architect certification for?
Google Cloud lists no formal prerequisites, but recommends 3+ years of hands-on experience building, testing, deploying and managing cloud solutions, including 1+ year building agentic solutions using Google Cloud.
Strong alignment
- Cloud architects and senior cloud engineers
- AI/ML engineers building production agent systems
- Developers moving into agentic application architecture
- Technical leads designing enterprise AI workflows
- Professionals comfortable with Google Cloud infrastructure and applications
May need foundation first
- People new to cloud architecture
- Learners with prompt-only experience and little coding
- Candidates without hands-on Google Cloud exposure
- People who know AI concepts but have not deployed cloud systems
- Beginners expecting a fundamentals-level certification
Important: no formal prerequisite does not mean beginner-level
“Prerequisites: None” describes registration eligibility. The recommended experience describes the depth of real-world knowledge Google Cloud expects the role to have.
How is this different from traditional cloud architecture?
Traditional cloud architecture remains important, but agentic systems introduce additional variables because system behavior is not always a fixed, deterministic sequence.
| Traditional architecture question | Agentic architecture adds another question |
|---|---|
| Where should the application run? | Where should the agent runtime run and what changes for long-running or tool-heavy workflows? |
| How does the application access data? | How does the agent retrieve authoritative information, preserve context and avoid poor retrieval? |
| How do services authenticate? | What identity should the agent use, what tools can it call and how is access bounded? |
| How do we test the application? | How do we evaluate an agent whose output can vary and whose tool choices affect the result? |
| How do we monitor services? | How do we spot reasoning loops, drift, hallucinations, latency and tool failures? |
| How do we secure the application? | How do we enforce guardrails, gateway controls and human oversight around autonomous actions? |
The architecture is no longer just the infrastructure around the AI. The agent’s behavior becomes part of the architecture.
What does the current beta status mean?
Google Cloud explains that beta certification exams are newly developed assessments. Performance statistics from beta questions are used to establish the standards for the final certification. Beta participants are notified of their results 4–6 weeks after both the exam window and lab window are closed.
Register for the current beta window
The Google Cloud certification catalogue currently states that the Professional Agentic Architect beta is open through September 30, 2026.
Complete the proctored assessment
The beta exam is currently listed at approximately 80 multiple-choice questions and 3 hours.
Complete the hands-on portion
After passing the multiple-choice beta, participants are automatically notified when the labs are available in Google Skills.
Wait for beta results
Google says beta participants receive results 4–6 weeks after both the exam and lab windows are closed.
Keep checking the live Google Cloud page
Google is updating certification exams to reflect products announced at Google Cloud Next ’26, including Gemini Enterprise Agent Platform. The exam guide remains the best source for the products currently in scope.
What should you understand before you start?
If you are exploring this certification, begin with the five capabilities Google Cloud is actually testing rather than memorizing the product list.
| Capability | What “understanding it” looks like |
|---|---|
| Agent construction | You can explain how a low-code or custom agent is structured and where prompts, skills, tools and state fit. |
| Enterprise knowledge | You can reason through RAG, embeddings, vector retrieval, reranking and enterprise data access. |
| Multi-agent orchestration | You understand handoffs, sequential and parallel workflows, A2A, MCP and agent identity. |
| Production engineering | You can compare runtimes, design evaluation, diagnose latency and failures, and optimize reliability and cost. |
| Security and governance | You can design access boundaries, gateway controls, policy enforcement, guardrails and human oversight. |
Quick self-check: does this sound like your current work?
This is a learning diagnostic, not a prediction of exam performance.
Building your Agentic Architect study path?
Use the official exam guide as the backbone, then add hands-on labs, architecture scenarios and practice questions that make you explain why one design fits a requirement better than another.
