Develop custom agents
Free Google Cloud Professional Agentic Architect practice — 6 questions on Develop custom agents, with explanations. No sign-up. Full 12-question mixed test →
A multi-agent ADK application deployed to Vertex AI Agent Engine must retain conversation state for a user across multiple days, surviving redeployments and scale-to-zero events. Which SessionService implementation should the architect configure?
VertexAiSessionService is the ADK-provided session service that stores sessions/events via the managed Agent Engine Sessions API, giving durable persistence across restarts and redeploys without custom infrastructure.
A customer support platform routes each inbound ticket to one top-level agent. For complex tickets, that agent must simultaneously invoke a billing sub-agent and a technical-support sub-agent, then merge both responses into one reply before returning to the user. Which multi-agent pattern does this describe?
A top-level orchestrator dynamically dispatches work to multiple worker sub-agents concurrently and then aggregates their outputs — this is the classic orchestrator-worker hierarchy using parallel invocation.
An external agent built on a non-ADK framework needs to discover the skills, authentication requirements, and endpoint URL of a remote ADK-based agent before sending it a task, using the A2A protocol. How does the remote agent expose this information?
The A2A protocol defines a standardized Agent Card, a JSON document published at a well-known path, that lets any A2A-compliant client discover capabilities, auth schemes, and the endpoint before submitting tasks.
During development, an architect wants to force a Gemini model inside an ADK agent to always emit a function call on every turn (never a plain text answer) so a downstream validator can inspect it before execution. Which configuration achieves this?
FunctionCallingConfig mode ANY forces the model to always return a function call, optionally constrained to an allowed_function_names list, guaranteeing tool-call output for the validator.
In a Vertex AI Agent Engine deployment, what is the key architectural difference between an agent's session state and its Memory Bank-managed long-term memory?
Session state is per-session, short-lived data tied to that session's lifecycle, while Memory Bank is a separate managed service that distills and stores durable, user-level facts and preferences retrievable across future, unrelated sessions.
An ADK agent must call tools exposed by an enterprise MCP server running on a separate machine, reachable only over the network rather than as a local subprocess. Which MCP transport configuration should the ADK MCPToolset use?
MCP supports remote connections via an HTTP-based transport using Server-Sent Events (or streamable HTTP), which is what ADK's MCPToolset should use to reach a network-hosted MCP server rather than a locally spawned one.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →