AI Field Guide

Orchestrator + isolated subagents for parallel or context-heavy work best practice multi-agent reviewed 1h ago
A lead agent holds the overall task and spawns short-lived, narrowly-scoped subagents, each with its own clean context window, to do parallelizable research, context-heavy lookups, or specialized work, then returns only a compressed summary to the orchestrator.
- Claude: When and how to build multi-agent systems
- Multi-Agent LLM Orchestration for Incident Response (arXiv)
Why this verdict?

Anthropic reports this winning specifically for parallelizable research, context isolation, and tool-set specialization -- their research system beat single-agent Claude by over 90% on their internal eval. But it's not a default: multi-agent setups use 3-10x the tokens of a single agent, and uncoordinated multi-agent setups have been shown to amplify errors far more than centrally-orchestrated ones. Anthropic's own advice is to start with one well-tooled agent and add subagents only once that proves insufficient.