Which two orchestration methods are MOST suitable for implementing complex agentic workflows that require both external data access and specialized…

For complex agentic workflows, a multi-faceted orchestration strategy is required. Retrieval-based orchestration (D) is essential for the "external data access" requirement. This method, commonly implemented via Retrieval-Augmented Generation (RAG), allows the agent to query external knowledge bases (e.g., documents, databases) and ground its responses in factual, up-to-date information. Agentic orchestration with specialized expert system delegation (A) addresses the "specialized task delegation" requirement. In this pattern, a primary orchestrator agent analyzes a complex request and routes sub-tasks to specialized agents (e.g., a data analysis agent, a code generation agent), effectively creating a "mixture of experts" to handle diverse tasks efficiently.

B. Prompt chaining is a low-level technique for passing context, not a high-level orchestration method for delegating tasks or accessing external data sources.

C. Manual coordination is inherently unscalable and unsuitable for complex, automated agentic workflows, defeating the purpose of using agents.

E. Static rule-based routing is too rigid for complex workflows that require dynamic, context-aware decision-making to select the appropriate tool or data source.

1. P. Lewis et al., "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks," Advances in Neural Information Processing Systems 33, 2020, pp. 9459-9474. This paper introduces RAG, the foundational technique for retrieval-based orchestration to access external data.

2. J. D. Fernández et al., "Building a Multi-Agent System with LangChain and GPT-4," IEEE Access, vol. 11, 2023, pp. 109235-109251, doi: 10.1109/ACCESS.2023.3320912. This publication discusses architectures where a "master" agent delegates tasks to specialized agents, demonstrating the "specialized expert system delegation" pattern.

3. NVIDIA AI Blueprints, "Conversational AI". These reference architectures often showcase patterns combining RAG for external knowledge with routing logic to different skills or models, embodying the principles of both correct answers for building robust systems.