Professional Agentic Architect Exam Guide: Five Domains, Weights
Google's Professional Agentic Architect exam guide gives custom agent development about 33% of the exam and evaluation and deployment 22%, across five sections.
By Dillip Chowdary • Nov 16, 2026 • Source: Google Cloud

At a glance
- Sections
- 5, weighted from about 13% to about 33%
- Largest section
- Developing custom agents (~33%)
- Second largest
- Evaluating and deploying agentic workflows (~22%)
- Tools in scope
- 28 Google Cloud products, frameworks and protocols
- Exam
- 2 hours, 50-60 multiple-choice questions, then 2 labs
In this post
Google Cloud's Professional Agentic Architect exam guide splits the new certification into five weighted sections, and developing custom agents in code carries the most weight at about a third of the exam. The guide also names 28 tools in scope, from the Agent Development Kit (ADK) to Model Armor, Antigravity and the Model Context Protocol.
This breakdown follows the guide section by section. For the launch details, fee and retake rules, see our report on the Professional Agentic Architect launch.
01Exam weights across the five PAA sections
The guide describes the certified architect as an experienced developer or architect who designs and manages autonomous, AI-driven agentic workflows in Google Cloud. It weights the five sections as follows.
| Section | Weight | Sub-areas |
|---|---|---|
| 1. Building agents using low-code tools | ~13% | Low-code workflows and behavior; connecting enterprise data to Gemini Enterprise |
| 2. Using coding agents for application development | ~17% | Using coding agents effectively; customizing them for enterprise workflows |
| 3. Developing custom agents | ~33% | Agentic workflows in code; enterprise domain knowledge; orchestration |
| 4. Evaluating and deploying agentic workflows | ~22% | Evaluation in development and production; deploying and scaling |
| 5. Securing and governing agentic workflows | ~15% | Security and governance; secure agent behavior and execution |
Sections 3 and 4 together make up about 55% of the exam. The weights apply to the multiple-choice part; the two hands-on labs that follow are a separate requirement, and the certification page says they validate execution and coding ability.
02Low-code agents and coding agents: sections 1 and 2
Section 1 tests configuring state-based workflows, with pages, transition routes and event handlers, in Gemini Enterprise tools such as Workflow Builder and Customer Experience Agent Studio (CX Agent Studio). It also covers writing system instructions and in-console prompt templates, including few-shot and chain-of-thought prompts.
The rest of section 1 is about data. Candidates are expected to connect agents securely to proprietary enterprise data through Gemini Enterprise and Agent Search, and to ingest unstructured multimodal data such as video, audio and images into an agentic workflow.
Section 2 is about coding agents themselves. The guide names configuring them with MCP servers, custom skills and tool access, with Antigravity and Claude Code on Google Cloud as examples, and running them in secure sandboxes on GKE, Cloud Workstations or Antigravity.
It also covers using coding agents to refactor code, optimize runtimes and patch application-layer vulnerabilities. Customization means creating skills, plugins, extension hooks, rules and subagents in Antigravity, and adding Agents CLI to build, scale, govern and optimize deployed agents.
03Developing custom agents: the 33% section
Section 3 starts with model choice. The guide asks candidates to pick between large and small language models, self-hosted and SaaS, and open-source and proprietary options, weighing cost, security and agent architecture.
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
The build topics include custom agents with open-source libraries such as ADK, sessions and memory with Agent Platform Memory Bank and managed sessions, and skills configured through Agents CLI, including plugins and agent versus human mode.
Enterprise knowledge comes next. The guide lists retrieval-augmented generation (RAG) pipelines and vector retrieval, covering embedding models, similarity scoring and reranking, on services such as Vector Search and Agent Retrieval, along with agent permissions through Agent Identity.
It also covers using Agent Registry and Google Cloud MCP servers to wire up prebuilt and custom capabilities, such as integration layers for managed databases, API integrations and MCP servers that reach third-party SaaS tools. Orchestration closes the section: agentic protocols such as MCP and Agent2Agent (A2A), plus parallel, sequential and graph multi-agent workflows using Agent Identity, Agent Registry, Agent Runtime and agent policies.
04Evaluation, deployment, security and governance
Section 4, at about 22%, begins with evaluation. Candidates should be able to build test sets from golden data, prompts and edge cases, run continuous evaluation pipelines on an agent's tool execution, and choose tooling such as ADK evalsets, the Agent Platform Gen AI evaluation service or custom autoraters.
The deployment half covers choosing a runtime among Agent Runtime, Cloud Run and GKE by use case, requirements and cost. It also covers troubleshooting drift, tool invocation latency, reasoning loops and system failures, and monitoring agents for logic errors, latency bottlenecks and hallucinations.
Section 5, at about 15%, covers authentication and secure tool execution, such as agent-to-tool API calls over OAuth 2.0, and Principal Access Boundary policies through Agent Identity. Candidates should also know how to use Agent Gateway to monitor traffic and track agents.
Governance rounds it out. The guide lists policy enforcement with Agent Registry and Model Armor, guardrails that use Agent Gateway, Model Armor and human-in-the-loop review, and secure data access with identity propagation.
05Tools in scope and official preparation resources
The guide lists 28 tools in scope for the exam. Several carry new names, which the guide notes: Agent Runtime was Agent Engine, Agent Search was Vertex AI Search, and Workflow Builder was Agent Designer.
- Agent Development Kit (ADK)
- Agent Gateway
- Agent Identity
- Agent Registry
- Agent Retrieval and Vector Search
- Agent Runtime (formerly Agent Engine)
- Agent Search (formerly Vertex AI Search)
- Agentic protocols (A2A, MCP)
- Agents CLI in Agent Platform
- Antigravity (CLI, SDK, App)
- BigQuery
- Cloud Run
- Cloud SQL
- Cloud Storage
- Firestore
- Gemini Enterprise
- Gemini LLMs
- Gen AI evaluation service
- Google Cloud Observability
- Google Kubernetes Engine (GKE)
- Memorystore for Redis
- Model Armor
- Model Context Protocol (MCP) servers
- Model Garden
- RAG Engine
- Sensitive Data Protection
- Skill Registry
- Workflow Builder (formerly Agent Designer)
Google points candidates to four free preparation resources.
- Read the exam guide. The official exam guide PDF lists every topic that may appear.
- Work through the learning path. The Google Skills learning path combines courses, labs and skill badges, and partners have a separate partner path.
- Try the sample questions. Google's sample question set shows the question format.
- Join GEAR. Gemini Agent Ready gives 35 free monthly Google Skills credits and agent codelabs.
Google's certification FAQ says the learning path is optional, and the labs taken after the exam are a distinct set not found in it. Employees of Google Cloud customers can prepare and earn a free voucher through the cohort described in our GEAR Special Edition application guide.
Frequently asked questions
What are the Professional Agentic Architect exam domains?
Building agents with low-code tools (~13%), using coding agents (~17%), developing custom agents (~33%), evaluating and deploying agentic workflows (~22%), and securing and governing them (~15%).
Which section of the PAA exam carries the most weight?
Developing custom agents, at about 33%. It covers model selection, ADK, memory, RAG and vector retrieval, agent permissions and multi-agent orchestration with MCP and A2A.
How many tools are in scope for the Professional Agentic Architect exam?
The exam guide lists 28, including ADK, Agent Runtime, Agent Gateway, Agent Identity, Antigravity, Gemini Enterprise, Model Armor, RAG Engine and Cloud Run.
Is the Professional Agentic Architect learning path required?
No. Google says the learning path is a preparation resource, and the two labs taken after passing the exam are a separate set.
Sources
- Google Cloud: Professional Agentic Architect exam guide (PDF)
- Google Cloud: Professional Agentic Architect certification page
- Google Cloud Certification Help: Professional Agentic Architect Certification FAQs
- Google Cloud Blog: Introducing the Professional Agentic Architect Certification
- Google Skills: Professional Agentic Architect learning path
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement