Google agentic architect beta: 90-day AI-agent skill test

Google Cloud opened beta registration for the Professional Agentic Architect certification on September 3, 2026, and closed it on September 30, 2026. If you are already shipping agent work, the credential can be useful proof. If you are still learning the shape of the work, use the same period to build the skill first.

The multiple-choice exam window ran from September 8, 2026 to September 30, 2026. Earning the Professional Agentic Architect credential required passing both the proctored multiple-choice exam and the hands-on labs. The beta is not a single test. That makes the window a planning constraint, not just a date on a calendar.

The beta exam fee was $120, a 40% discount from the $200 USD retail price. The generally available Professional Agentic Architect certification will cost $200 USD. Price is rarely the deciding factor. Readiness is, because the exam asks you to show the work, not just recognize the terms.

Google recommended at least 3 years of hands-on cloud experience and at least 1 year building agentic solutions on Google Cloud for the Professional Agentic Architect exam. If your background is shorter, the exam is likely to expose gaps you would rather find in a project. Treat that as a planning signal: choose the path that makes the gap visible before the deadline.

US job listings mentioning agentic AI skills rose from 0.06% in 2024 to 0.23% in 2025, a year-over-year jump of more than 280%, or nearly 90,000 postings. The market is moving, but a badge does not create the underlying ability. It can make that ability easier to see when you already have it.

The beta is proof, not a shortcut

A certification works best when it confirms what you can already do. It should not be the place where you learn the job. Explain an agent design, connect a model to tools, test outputs, and keep the system under control, and the credential gives that ability a visible label. Struggle with any of those, and the exam will feel like a speed run through a skill you have not built.

Employers are not just looking for a badge. They are looking for people who can ship agent work without turning it into a demo. The beta is a chance to stand out, but only if your portfolio can back it up. A strong candidate can point to a small system that handles a real task, fails safely, and can be reviewed by another engineer.

The readiness checklist is the decision point

Use this list as a gate. Each item is meant to be checked quickly, not as a full audit. Hesitation on an item is the answer. If you can check all five, register. If you cannot, build one end-to-end agent project first.

• Sketch an agent loop: trigger, context, tool calls, stop condition, and failure path.
• Wire a large language model to a real task, not just a demo prompt.
• Define pass/fail tests and run them against sample inputs.
• Name the guardrails: permissions, data boundaries, logging, and human review.
• Explain how the agent talks to tools and other agents, including MCP/A2A, the connective patterns for agent work.

Design separates people who have used a chatbot from people who have built an agent. Integration checks whether you can move from prompt to real work. Testing is the difference between a toy and something you would trust in production. Governance is where many candidates get stuck, because guardrails are not a nice-to-have. Connection keeps you from building an agent that cannot talk to the wider system.

The missing piece can be built before the window ends

Pass the list? Register and prepare with the same discipline you would use for a project. Fail it? Do not spend the time memorizing exam material. Pick the weakest item and build a small agent that proves it. A simple support triage agent, a document question-answering agent, or a workflow automation agent can cover the gap if you keep the scope tight.

This week, pick the task, write down the checklist results, and schedule the build. Start with a clear input, output, and success measure; add a model call, a tool, and a stop condition. If any gap is found, make the build the priority. Finish the project before the beta window ends, and you can register with confidence. Leave it unfinished, and you still have a stronger skill set and a clearer path to the credential later.