Google Professional Agentic Architect: beta closed, exam opens in mid-November

The full title is Google Cloud Certified - Professional Agentic Architect. It targets developers, cloud architects and engineers who build autonomous AI agent workflows on Google Cloud with an eye on reliability, cost and security. Theory alone will not do: the official profile asks for experience with large language models, agent design patterns, code and data source integration.

There are no prerequisites, but Google recommends 3+ years of hands-on experience building, testing, deploying and managing cloud solutions, including 1+ years building agents on Google Cloud.

It is a hybrid certification. The multiple-choice exam, delivered by Pearson, tests conceptual knowledge and design choices. If you pass, you move on to labs on Google Skills that check you can execute and code in a live environment. Passing the exam alone does not get you certified.

The beta had registration from 3 to 30 September 2026 and the exam from 8 to 30 September. It cost USD 120, 40% less than the USD 200 final price. Google set the multiple-choice results for the end of October. Those who passed have until 31 December 2026 to complete the labs, estimated at under 5 hours, and the exam page places the final result 4-6 weeks after both windows close.

The general version keeps both parts and changes the rest: 2 hours instead of 3, and an exam result in 7-10 business days instead of weeks of waiting. You then have 2 months to complete the labs. Registration also opens in mid-November. Google does not yet publish the exact day, the number of questions or the languages of the general version, or the passing score.

Google justifies the 1-year validity by how fast software changes and announces a continuous education path to keep the certification active.

The [official guide](https://services.google.com/fh/files/misc/professional_agentic_architect_exam_guide_english.pdf) splits the content into five sections. The heaviest, at ~33%, is developing custom agents: choosing a model, building with Agent Development Kit (ADK), configuring sessions and memory with Memory Bank, setting up RAG pipelines with Vector Search and Agent Retrieval, granting permissions with Agent Identity, and coordinating several agents with MCP and Agent2Agent (A2A), in parallel, in sequence or as a graph.

Next comes evaluating and deploying, at ~22%: test sets with golden data and edge cases, continuous evaluation pipelines, ADK evaluation tooling, choosing between Agent Runtime, Cloud Run and GKE, and troubleshooting drift, tool invocation latency and agent reasoning loops. These two sections add up to 55% of the exam.

Using coding agents weighs ~17%: configuring them with MCP servers and custom skills (the guide names Antigravity and Claude Code on Google Cloud) and running them in secure sandboxes. Securing and governing agents weighs ~15%: OAuth 2.0 for tool calls, Agent Gateway, Agent Registry, Model Armor and human-in-the-loop controls. Low-code tools close the list at ~13%: Gemini Enterprise Agent Designer and CX Agent Studio.

Watch the names when you study from older material: the guide uses Agent Runtime for what used to be Agent Engine, and Agent Search for what used to be Vertex AI Search.

If you passed the beta exam, finish the labs by 31 December 2026. If you failed, you can try again when the general version opens, and the beta attempt does not count toward the allowed total.

If you are starting now, read the guide and the [official exam page](https://cloud.google.com/learn/certification/agentic-architect), follow the [Google Skills learning path](https://www.skills.google/paths/4525), which is optional, and answer the sample questions linked from the exam page. Then build an agent with ADK, evaluate it against a golden dataset, deploy it to Agent Runtime or Cloud Run, and protect it with Model Armor. That run touches the heaviest sections; the certification labs are distinct from the ones in the learning path.

If you lack Google Cloud architecture grounding, start there. The Professional Cloud Architect guide asks you to map compute needs to products such as GKE and Cloud Run, to secure AI with Model Armor and Sensitive Data Protection, and to know monitoring and logging; all of that comes back in Agentic Architect. With CertifyDeck's Professional Cloud Architect questions you can check that level, and you can start free. That guide does not include ADK, MCP, A2A or Agent Gateway, though: you will have to study those separately.

If you come from the business side, the Generative AI Leader questions go over RAG and techniques such as few-shot and chain-of-thought, which also appear in this guide. But it centres on strategic leadership, not technical implementation: it only gets you the vocabulary.