Agents and Workflows
Free Claude Certified Developer (CCDV-F) practice — 6 questions on Agents and Workflows, with explanations. No sign-up. Full 12-question mixed test →
A financial services company is building an agent that must perform multi-step loan approval workflows. The agent needs to call external credit scoring APIs, validate documents using vision models, query internal databases for compliance rules, and generate approval letters. The workflow must support conditional branching based on credit scores and allow human-in-the-loop review for edge cases. Which solution BEST meets these requirements?
Agents with tools provide native multi-step orchestration, automatic tool selection based on context, built-in conversation memory for workflow state, and support for complex conditional logic. The human-in-the-loop requirement is naturally handled through agent conversation flow. This is the intended use case for the Agents API.
An e-commerce company's customer service agent uses Tools to check inventory, process returns, and update order status. During testing, the agent sometimes calls the wrong tool or uses incorrect parameters when customers phrase requests ambiguously. The team needs to improve tool selection accuracy without rewriting tool definitions. What is the MOST effective approach?
System prompt enhancement with concrete examples is the most effective way to improve tool selection accuracy. Providing specific customer phrasing patterns (e.g., 'I want to send this back' → use return_tool) directly guides the model's tool selection logic. This leverages few-shot learning within the agent's instruction set.
A research team is building an agent that analyzes academic papers by extracting citations, summarizing methodology sections, and comparing results across multiple papers. A single analysis workflow processes 50 papers averaging 30 pages each. The agent uses vision capabilities to extract tables and figures. The team is concerned about context window utilization and cost. Which optimization strategy provides the BEST balance of accuracy and efficiency?
Batch API is specifically designed for high-volume, parallelizable tasks where results aren't needed immediately. Processing papers independently avoids context window bloat (50 papers × 30 pages would exceed limits), provides 50% cost reduction, and allows parallel processing. Client-side aggregation is appropriate since papers are independent analysis units.
A healthcare agent processes patient intake forms using vision capabilities to extract handwritten notes and checkboxes. The workflow must validate extracted data against HL7 FHIR standards, flag missing required fields, and route cases requiring specialist review. Compliance rules require that all data extraction decisions are auditable. Which implementation approach BEST satisfies these requirements?
Agents provide built-in conversation logging that captures the full decision flow including tool calls, tool responses, and reasoning. Separating extraction (vision tool) from validation (FHIR tool) creates clear audit points. The agent's orchestration naturally handles the routing logic based on validation results, and conversation logs satisfy auditability requirements.
An agent-based code review system analyzes pull requests by examining changed files, running static analysis tools, checking style guides, and posting review comments. The system processes 200 PRs daily with an average of 12 files changed per PR. To optimize costs while maintaining review quality, the team wants to cache common elements. What is the MOST cost-effective caching strategy?
System prompt caching (5-minute minimum TTL) for style guides and tool definitions provides 90% cost reduction on cached tokens and perfectly fits this use case—style guides are static and used across all PRs. Processing PRs independently avoids context pollution and allows parallel processing. File diffs are PR-specific and shouldn't be cached.
A legal document analysis agent must process contracts by extracting clauses, identifying risks, cross-referencing with case law databases, and generating summaries. The agent uses 8 different tools including database queries, PDF extraction, and risk scoring APIs. During peak usage, the agent sometimes exceeds rate limits on individual tools. The team needs to handle tool failures gracefully without losing workflow progress. Which approach provides the MOST robust error handling?
Tool-level retry logic with error details returned to the agent allows the agent to receive actionable information about failures and make intelligent decisions (use alternative tools, skip optional steps, request human intervention). This keeps the agent in control of workflow logic while tools handle their own resilience. The agent can adapt its strategy based on specific error types.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$54.99 $17.99 with code FREETEST33 — valid through September 1.
Get my $17.99 deal →