Professional Agentic Architect exam guide, section by section

Section 1: Building agents using low-code tools (about 13%)

Section 1 is about 13% of the exam and lists four considerations under 1.1 (low-code workflows and instructions) and 1.2 (enterprise data in Gemini Enterprise). Most questions turn on three decisions: whether a rule belongs in instructions or in a deterministic control such as a flow route, handoff rule or Condition step; whether a data source is ingested or federated and which identity setup it needs; and how PDFs, video and audio become retrievable text.

Section 1 study page: decision tables, commonly confused products and 5 free questions with answers.

Section 2: Using coding agents for application development (about 17%)

Section 2 is about 17% of the exam and lists five considerations under 2.1 (using coding agents) and 2.2 (customizing them for enterprise workflows). Most questions turn on where a setting lives (a workspace file committed to the repository, a global file on one machine, or a managed setting or IAM policy enforced centrally), which customization is deterministic (hooks, IAM deny policies, managed settings) rather than model-dependent (rules, skills), and which sandbox setting keeps agent commands isolated while still reaching what they need.

Section 2 study page: decision tables, commonly confused products and 7 free questions with answers.

Section 3: Developing custom agents (about 33%)

Section 3 is about 33% of the exam and lists nine considerations across 3.1, 3.2 and 3.3. Most questions turn on where a control belongs: in code or configuration (callbacks, plugins, workflow agents, graph nodes, IAM and deny policies), not in the model's instructions. The rest turn on which identity holds a permission (the agent's own identity, not a developer account or a shared service account) and which protocol fits the boundary (MCP for tools, A2A for agents owned by others).

Section 3 study page: decision tables, commonly confused products and 13 free questions with answers.

Section 4: Evaluating and deploying agentic workflows (about 22%)

Section 4 is about 22% of the exam and lists seven considerations: four under 4.1 (evaluation) and three under 4.2 (deployment and scaling). Most questions turn on three choices: which evaluation tool and metric fits the data you have (reference answer or rubrics, LLM judge or deterministic, offline evalset or online monitor), whether the agent belongs on Agent Runtime, a Cloud Run resource type or GKE, and which setting or observability view fixes a production symptom such as drift, slow tools or a reasoning loop.

Section 4 study page: decision tables, commonly confused products and 9 free questions with answers.

Section 5: Securing and governing agentic workflows (about 15%)

Section 5 is about 15% of the exam and lists six considerations across 5.1 (configuring security and governance) and 5.2 (secure agent behavior and execution). Most questions turn on three choices: which Agent Identity credential model fits the call (3-legged OAuth, 2-legged OAuth, OIDC ID token or API key), where a control must live (Principal Access Boundary on a principal set, Unified Access Policy on the egress gateway, Model Armor on ingress, egress or inside the agent), and the Agent Gateway placement and hostname allowlisting rules that decide whether agent traffic is allowed at all.

Section 5 study page: decision tables, commonly confused products and 6 free questions with answers.

Tools in scope, with their former names

The current guide uses the 2026 names. Copies of the guide from the beta period name Agent Designer in Section 1, the product now called Workflow Builder; older courses also use Agent Engine and Vertex AI Search. When a study resource uses a name in the "Formerly" column, it describes the same product.