Agent Atlas

Atlas · agents

Subagents

A helper agent that works in a separate context and returns its results to the main agent.

Established practiceDepth 3 of 4Reviewed 2026-11-14

In plain words

A subagent is a second agent the main one starts for a side task, like searching a codebase. It does the messy work in its own context and returns a short answer, so the main conversation stays clean.

How it works

Claude Code runs each subagent in its own context window, and each can have its own tools and permissions. A fresh subagent knows only what its task prompt tells it, so the main agent has to hand over what matters.

Going deeper

Subagents keep the main window clean, but they do not save tokens: each one gathers its own context, and Anthropic measured its multi-agent research at about 15 times the tokens of a chat. Use them for parallel work and a fresh pair of eyes. They are one answer to context rot. Others include compaction, clearing between tasks, and notes kept in files.

Sources

  1. Create custom subagents Anthropic
    • Each subagent runs in its own context window
    • Each subagent starts with a fresh, isolated context window.
    • Claude composes a delegation message that summarizes the task, and the subagent works from there.
  2. Effective context engineering for AI agents Anthropic
    • Context refers to the set of tokens included when sampling from a large-language model (LLM).
    • the model’s ability to accurately recall information from that context decreases.
    • Context, therefore, must be treated as a finite resource with diminishing marginal returns.
    • This results in n² pairwise relationships for n tokens.
    • Compaction typically serves as the first lever in context engineering to drive better long-term coherence.
    • These notes get pulled back into the context window at later times.
    • but returns only a condensed, distilled summary of its work
  3. How we built our multi-agent research system Anthropic
    • Our Research system uses a multi-agent architecture with an orchestrator-worker pattern, where a lead agent coordinates the process while delegating to specialized subagents that operate in parallel.
    • Each subagent needs an objective, an output format, guidance on the tools and sources to use, and clear task boundaries.
    • multi-agent systems use about 15× more tokens than chats.

Quotes were checked against the live pages on 2026-11-08.