What you'll learn:
- Build AI agents on Google Cloud two ways — low-code with Gemini Enterprise, and in code with the Agent Development Kit (ADK)
- Design production RAG pipelines: embeddings, chunking, similarity scoring, reranking, Vector Search and Agent Retrieval
- Orchestrate multi-agent systems using MCP and Agent2Agent (A2A), with sequential, parallel and graph workflows
- Evaluate agents properly — golden datasets, trajectory vs. response, ADK evalset, autoraters and continuous evaluation
- Deploy and scale agents on Agent Runtime, Cloud Run and GKE, and choose correctly between them
- Secure agentic workflows against prompt injection using Model Armor, Agent Gateway, Agent Identity and PAB policies
- Use coding agents effectively — Antigravity, Claude Code on Google Cloud, MCP servers, custom skills and secure sandboxes
- Configure agent memory and state with managed sessions and Agent Platform Memory Bank
- Troubleshoot real agent failures: drift, reasoning loops, tool latency, logic errors and hallucinations
- Cover 100% of the Professional Agentic Architect exam guide, section by section, in Google's own order
Everything on the Google Cloud Professional Agentic Architect exam — and nothing that isn't.
This course covers all five domains of the Professional Agentic Architect certification, in the exact order Google's own exam guide presents them. Every objective is covered. Every in-scope tool is explained.
Structured to match the exam, not to pad the runtime
Most certification courses spend roughly equal time on every topic. This one doesn't, because the exam doesn't:
Section 1 — Building agents using low-code tools · 13% of the exam · 6 lectures
Section 2 — Using coding agents for application development · 17% · 7 lectures
Section 3 — Developing custom agents · 33% · 14 lectures
Section 4 — Evaluating and deploying agentic workflows · 22% · 9 lectures
Section 5 — Securing and governing agentic workflows · 15% · 6 lectures
Forty-two lectures, each sized to what it's actually worth. Section 3 is a third of the exam, so it gets a third of the course.
One worked example, from start to finish
Instead of five disconnected demos, a single company — Sunrise Electronics — runs through the entire course.
You'll watch them build a low-code support agent, hand tools to a coding agent, write a custom warranty agent in ADK, connect it to their own documents with RAG, prove it actually works, deploy it to production, and finally survive a security review.
By the end, you haven't just learned a list of services. You've seen one real system grow from nothing to production.
What you'll actually cover
Building agents (low-code): Gemini Enterprise, Agent Designer, CX Agent Studio, pages and transition routes, event handlers and parameters, system instructions, few-shot and chain-of-thought prompting, Agent Search, and multimodal ingestion of images, audio and video.
Coding agents: Antigravity (CLI, SDK and App), Claude Code on Google Cloud, MCP servers, custom skills and the Skill Registry, secure sandboxes on Cloud Workstations and GKE, plus Agents CLI for operating a whole fleet.
Custom agents in code: Agent design patterns, LLM vs. SLM selection, Model Garden and Gemini models, the Agent Development Kit (ADK), tools and function calling, callbacks, managed sessions, Agent Platform Memory Bank, RAG end to end — embeddings, chunking, similarity scoring, reranking — Vector Search, Agent Retrieval, RAG Engine, Agent Identity, Agent Registry, the MCP and A2A protocols, and multi-agent orchestration.
Evaluating and deploying: Why agent testing is nothing like normal software testing, trajectory vs. final response, golden datasets, response and retrieval quality metrics, the ADK evalset, the Gen AI evaluation service, custom autoraters, continuous evaluation pipelines, choosing between Agent Runtime, Cloud Run and GKE, Cloud Logging and Cloud Trace, and troubleshooting drift, reasoning loops, latency and hallucinations.
Securing and governing: The agent threat model, prompt injection, OAuth 2.0 and Auth Manager, principal access boundary (PAB) policies, Agent Gateway, Model Armor, Agent Registry governance, human-in-the-loop patterns, and Sensitive Data Protection.
Taught in plain English
Every abstract idea gets a concrete anchor.
MCP is explained as USB for agents — one standard plug instead of a different cable for everything. Sandboxing is giving the agent a workshop, not the run of the building. A principal access boundary is the wall around the building, where IAM is just the key you were handed.
Nothing is used before it's defined, and no lecture runs longer than eight minutes.
Is this course for you?
This is a professional-level certification. Google's own exam guide describes the ideal candidate as an experienced developer or architect who can write code and has worked with large language models.
You'll get the most from this course if you're comfortable with Python, know your way around Google Cloud basics like IAM and Cloud Storage, and understand roughly how APIs work.
You do not need any prior experience building agents. Every concept is built up from first principles.
Ready to start?
The agentic architect certification is new, and the skills it tests — ADK, RAG, MCP, A2A, agent evaluation, agent security — are the ones every cloud team is scrambling to build right now.
Enroll today and start with Section 1.
