Delegation pattern enables main agents to assign subtasks to specialized sub-agents. Mechanism: Primary agent analyzes task, identifies subtasks requiring specialized skills, delegates to appropriate sub-agents, integrates results. When to delegate: Task requires specialized knowledge, subtask is well-defined, efficiency gain from specialization, reduces cognitive load on main agent. Implementation: Query router → specialist selection → context preparation → delegation call → result integration. Specialist types: Domain experts (legal, medical), tool specialists (code, web search), format experts (summarization, translation). Context management: Pass relevant context, not full conversation, minimize token usage, handle confidentiality. Return protocols: Structured results, confidence scores, error handling, partial results. Delegation criteria: Skill match, availability, cost/latency trade-offs. Frameworks: LangChain tool wrappers, CrewAI delegation, custom routing logic. Best practices: Clear task descriptions, verify delegation success, handle specialist failures, avoid infinite recursion. Anti-patterns: Over-delegation (everything needs specialist), under-delegation (monolithic agents). Effective delegation is key to scalable multi-agent architectures.