Build agents using low-code tools
Free Google Cloud Professional Agentic Architect practice — 6 questions on Build agents using low-code tools, with explanations. No sign-up. Full 12-question mixed test →
A retail company is building a customer service agent in Vertex AI Conversational Agents. The agent must handle open-ended questions about store policies using generative responses, but must enforce that any refund over $500 strictly follows a deterministic approval sequence (verify order, check manager override flag, log to compliance system) with no steps skipped or reordered. Which approach BEST meets these requirements?
Flows provide deterministic, guaranteed-order state-machine execution, while Playbooks handle flexible generative conversation. Conversational Agents supports invoking a Flow as a tool from within a Playbook, combining generative flexibility with strict, auditable execution for compliance-critical steps.
A company needs an agent that answers employee questions by grounding responses in a continuously updated repository of 50,000 internal PDF policy documents, with citation links to source documents, using Google Cloud's low-code tooling. Which approach requires the LEAST custom code?
Vertex AI Search data stores natively handle unstructured document ingestion, chunking, embeddings, and nightly re-indexing, and provide grounded answers with citations out of the box; attaching this as a Data Store agent/tool requires no custom RAG pipeline code.
An architect is building a Playbook-based agent in Vertex AI Conversational Agents that must never reveal internal system prompts or recommend competitor products, even under prompt injection attempts. Which low-code configuration provides the STRONGEST protection, without writing custom orchestration code?
Explicit negative/guardrail instructions combined with least-privilege scoping of tools and data stores is the recommended low-code defense-in-depth approach for Playbooks: it directly constrains disclosure behavior and limits the blast radius even if instructions are partially bypassed.
A team is using Vertex AI Agent Builder Playbooks to orchestrate order fulfillment: reserve inventory, charge payment, schedule shipment. If shipment scheduling fails, the agent must automatically execute compensating transactions (release inventory, refund payment) with full programmatic control over retry logic, transaction state, and error handling across services. Which statement BEST describes the right tooling choice?
Reliable multi-step transactional orchestration with programmatic retry, state tracking, and compensation logic across backend services exceeds what declarative low-code Playbooks/Flows guarantee. The Agent Development Kit (ADK) is the code-first framework designed for this, and a low-code agent can still call this logic as a tool — reflecting the documented limits of managed low-code builders.
A company wants employees to search and chat across Google Drive, Confluence, Jira, and ServiceNow content using natural language, without engineers building any conversation flows, intents, or playbooks. Which Google Cloud offering is purpose-built for this with pre-built connectors, requiring the LEAST low-code configuration?
Agentspace/Gemini Enterprise is purpose-built for enterprise search and chat experiences with pre-built connectors to common SaaS applications (Drive, Confluence, Jira, ServiceNow, etc.), requiring minimal configuration and no custom flows or playbooks.
A Conversational Agents Data Store agent grounded on a company's product manuals is giving confident-sounding but incorrect answers when the data store doesn't actually contain relevant information for a user's query. Which low-code configuration change should the architect make to reduce this behavior?
Data Store agents expose a configurable confidence threshold for grounded answers. Raising the threshold for what counts as sufficient grounding and defining an explicit fallback response for low-confidence results reduces hallucinated-but-confident answers when the source documents don't actually cover the query.
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 →