Skip to content

Definition

Delegating to a subagent by specifying goals and quality criteria, not rigid step-by-step procedures — letting the specialist choose its own strategy.

Key points

  • The trap: giving a web-search subagent detailed step-by-step procedural instructions causes it to fail rigidly on emerging topics or miss tangential sources.
  • Procedural Micromanagement (anti-pattern): Step 1: Search X → Step 2: Read Y → Step 3: Extract Z → Failure/Missed Value (rigid, not adaptable).
  • Goal-Oriented Delegation: the coordinator gives the subagent a Target (Coverage Breadth) & Criteria (Recency) → adaptable, high-value results (self-directed strategy).
  • The approach: specify research goals and quality criteria rather than procedural steps; let the specialized subagent determine its own search strategy. Keep tool interfaces generic but add enum parameters to guide behavior.
    • Example tool: tool: 'analyze_document', params: {analysis_type: 'extraction' | 'summarization'}.
  • Complements Structured Intermediate Representations and Shared Memory Architecture in multi-agent designs.

Why it matters for the exam

  • Multi-Agent Research scenarios test delegation style: goals+criteria beat procedural micromanagement; enum params beat brittle prose steps.

Common gotchas

  • Over-specifying procedure makes the subagent brittle — specify the what, not the how.
  • Guide behavior via enum tool params (extraction | summarization), keeping interfaces generic.

See also

Sources