Google Cloud has opened Professional Agentic Architect to the people who design, build, and ship agents on its platform. The company already put this curriculum in front of thousands of its own forward-deployed engineers and strategic partners. That training is now a public certification you can sit.
This isn’t a badge for writing clever prompts. Google describes the role as a developer or architect who designs and manages autonomous, AI-driven agent workflows in Google Cloud, and who has to weigh reliability, performance, cost, security, and scale. You show that in two stages: a proctored written exam, then a separate set of hands-on labs in a live cloud environment. You need a pass on both before you get the credential and the digital badge. Clearing the multiple-choice paper on its own doesn’t do it.
Introducing the new Professional Agentic Architect certification! Proving your expertise in building production-ready agents just got easier.
November 11, 2026
Used to train thousands of our own FDEs and partners, this certification is now available to all technical professionals who architect, build, and deploy agents on Google Cloud.
Check out more in our blog →
How the hybrid exam works
A useful way to picture it is a practising licence, not a quiz. You sit a design exam first. If you pass, you’re asked into a workshop to build, test, and debug a live system. The certificate shows up only after you’ve done both.
Part 1 is a proctored multiple-choice exam, delivered by Pearson. The generally available paper runs for two hours, in English, and has 50 to 60 questions. You can sit it from home with an online proctor, or at a test centre. The paper tests conceptual knowledge, system design, architectural standards, cost optimisation, and security governance. Official results take 7 to 10 business days.
Pass that, and CM Connect — Google Cloud’s certification portal — tells you you’ve reached Part 2. You then have two months to finish a distinct set of hands-on labs on Google Skills. These aren’t the labs from the optional learning path. Only people who’ve already passed the written exam get them, and they run in a live cloud project, so you have to build, test, and debug real agent workflows.
Who it’s for
Google is aiming this at developers, cloud architects, and engineers. There are no formal prerequisites, but the recommended bar is high: three or more years of hands-on work building, testing, deploying, and managing cloud solutions, and at least a year of that spent building agents on Google Cloud. The exam guide also assumes you’re already comfortable with large language models, agent design patterns, writing code, and hooking agents up to data in Google Cloud.
There’s an optional learning path on Google Skills, and a partner version on the partner Skills site. Google calls the path a starting point, not a requirement, and says you should already be comfortable with everything in the official exam guide. The path listing still carries a note that it will be updated for general availability, plus a separate warning about intermittent trouble with Gemini Enterprise labs.
What’s on the exam
The official exam guide cuts the written paper into five domains. Custom agent development is the heavy one, at about a third of the questions. Low-code tools take the smallest share.
| Domain | Share of the written paper |
|---|---|
| Low-code tools | ~13% |
| Coding agents | ~17% |
| Custom agents | ~33% |
| Evaluate and deploy | ~22% |
| Secure and govern | ~15% |
Domain one is the low-code corner. You’ll see Gemini Enterprise Workflow Builder, which used to be called Agent Designer, plus Customer Experience Agent Studio, system instructions and in-console prompt templates, Agent Search (the product previously known as Vertex AI Search), and work with multimodal unstructured data.
Domain two is about using coding agents to write software. The guide puts Model Context Protocol (MCP) servers and custom skills on the table, along with Google’s Antigravity — the CLI, the SDK, and the app — and Claude Code on Google Cloud. You’re also expected to know secure sandboxes on Google Kubernetes Engine, Cloud Workstations, and Antigravity; how to refactor, optimise, and patch existing code; and Antigravity’s skills, plugins, hooks, rules, and subagents, plus Agents CLI.
Custom agents account for about 33 percent of the paper, so this is where most of the marks sit. You may have to choose a large language model or a smaller one, a self-hosted setup or a software-as-a-service one, and an open-source model or a proprietary one, with cost, security, and architecture in mind. The same domain takes in the Agent Development Kit (ADK), Agent Platform Memory Bank and managed sessions, and Agents CLI skills. It also covers retrieval-augmented generation with Vector Search and Agent Retrieval, Agent Identity and Agent Registry, Google Cloud MCP servers, and orchestration over MCP and the Agent2Agent (A2A) protocol, including parallel, sequential, and graph workflows. Agent Runtime, formerly Agent Engine, and agent policies are in scope here too.
Evaluating and deploying those workflows is about 22 percent of the paper. That means test sets and continuous evaluation, ADK evalsets, the Agent Platform Gen AI evaluation service, custom autoraters, and golden datasets, plus a choice of runtime — Agent Runtime, Cloud Run, or GKE. You’ll also need to chase down drift, latency, reasoning loops, and failures, and to keep an eye on performance, reliability, and cost, including hallucinations.
Security and governance take about 15 percent. The guide calls out OAuth 2.0 for agent-to-tool calls, Principal Access Boundary policies paired with Agent Identity, Agent Gateway, Agent Registry, Model Armor, human-in-the-loop controls, and identity propagation.
The in-scope list doesn’t stop there. The guide also names BigQuery, Cloud SQL, Cloud Storage, Firestore, Memorystore for Redis, Gemini Enterprise, Gemini models, Model Garden, RAG Engine, Sensitive Data Protection, Skill Registry, and Google Cloud Observability, meaning Cloud Logging and Cloud Trace. Several products have been renamed, so if you learned the stack under an older title, trust the current exam guide rather than an old blog post.
Price, language, and a one-year credential
The generally available exam costs US$200, plus tax where it applies, which Google says matches its standard professional-level fee. It’s offered in English only. You can sit the written paper online with a proctor or at a Pearson site.
The credential is valid for one year. That’s shorter than other Google Cloud professional certifications, which are typically valid for two. Google’s official reason is the pace of software and algorithm change. The company says the certification has a continuous education path to help you keep an active status; the public exam page and FAQ don’t spell out the exact maintenance steps.
Google first ran a beta, with a longer paper and a discounted fee. The generally available exam uses the standard two-hour professional format and the standard fee. Register and read the official details on the Professional Agentic Architect exam page. The exam guide and the certification FAQ are the definitive sources for scope, format, and rules.
Where the curriculum came from
In a Cloud blog post, Gary Eimerman, managing director of Google Cloud Learning, and Ravi Rajamani, vice president of Applied AI and forward-deployed engineers, describe the certification as the latest milestone in Google’s Gemini Agent Ready, or GEAR, programme. They say nearly one million developers are using GEAR to skill up on agents. Over the past four months, they write, Google Cloud trained thousands of forward-deployed engineers and strategic partners on an advanced curriculum that began as an internal training initiative and is now open to the wider developer and partner community. A companion post on blog.google covers the same launch.
Google also cites research it ran with Ipsos on AI skill building. In the Cloud blog, 82 percent of learners using Google Skills say mentioning Google Cloud certifications on a résumé draws more recruiter responses, and 90 percent of cloud professionals say credentials like Google Cloud certifications give them the verified know-how to fill in-demand roles. The blog.google post adds that more than 80 percent of Google Cloud certified individuals say the credential opens doors to new career opportunities and speeds their path to promotion, and 82 percent say Google Cloud certifications help them earn higher salaries. Those figures are from Google’s survey with Ipsos, not an independent audit of hiring outcomes.
When Google Cloud opened beta registration, Richard Seroter, senior director and chief evangelist at Google Cloud, pointed at the new job title and the new exam:
Who knows exactly what jobs will be most relevant in a few years.
September 3, 2026
"Agentic architect" may be one. @googlecloud just launched a new beta certification for Professional Agentic Architect. Registration opens today.
Looks solid!
How it compares
Microsoft already offers a role-based expert credential in the same neighbourhood: Microsoft Certified: Agentic AI Business Solutions Architect Expert. To earn it you have to pass exam AB-100 and hold at least one listed associate-level certification, from a roster that includes Azure AI Apps and Agents Developer Associate and AI Agent Builder Associate. The Microsoft stack in the official description is Foundry, Copilot Studio, Power Platform, and Dynamics 365. It also covers A2A and MCP, multi-agent orchestration, and Microsoft’s responsible AI standard. The credential page doesn’t publish a single global exam fee; price is set by the country or region where the exam is proctored. Microsoft says its role-based certifications expire unless you renew them, and that renewal is a free online assessment on Microsoft Learn. AB-100 is offered in many more languages than Google’s English-only paper.
AWS doesn’t publish a dedicated “agent architect” certification. The closest official option is AWS Certified Generative AI Developer – Professional (AIP-C01). That exam costs US$300, lasts 180 minutes, and has 75 multiple-choice or multiple-response questions. It’s offered in English, Japanese, Korean, and Simplified Chinese, online or at a Pearson VUE centre. The official page describes it as a credential for developers who build and deploy production-ready AI solutions with AWS services such as Amazon Bedrock. An AWS Training and Certification blog post says the standard exam has been refreshed to include Amazon Bedrock AgentCore. It’s a written professional exam, not a hybrid of a paper plus a required post-exam lab in the way Google Cloud has designed Professional Agentic Architect.
Limits and caveats
This is a Google Cloud credential. The tools, runtimes, identity model, and evaluation services on the exam are Google Cloud’s. A pass won’t, on its own, speak to how you’d design the same system on another cloud.
The one-year validity is shorter than Google Cloud’s usual professional cycle, so you’ll be maintaining this one more often. The exam is English only. The recommended experience assumes you’ve already spent a year shipping agents on Google Cloud, even though there is no formal prerequisite. And because custom agent development is the largest domain, fluency in a low-code console isn’t enough: the paper also expects you to choose models, wire memory and retrieval, orchestrate agents over MCP and A2A, and reason about runtime, policy, and cost.
The two-month lab window starts after you pass the written exam, so a slow start can leave you short of time. The optional learning path is useful orientation, not a complete syllabus, and the path page still carries Google’s own notes about a pending general-availability update and intermittent Gemini Enterprise lab issues. For the live list of what can appear on the paper, use the official exam guide.
Discover more from Sulat AI & Tech News
Subscribe to get the latest posts sent to your email.
