WTF is Subagents?
Delegating a narrow task to a helper agent with its own context, so only a summary comes back to the main agent.
Also called: subagent, sub-agents, subagents for context management
TL;DR
Exploring a codebase or reading 40 search results fills an agent's context with noise. Hand that job to a subagent: it does the messy work in a fresh context and returns a short answer. The main agent stays focused, and the subagent can use a cheaper model.
How it works
Subagents absorb exploration, tool output, retries and intermediate reasoning; the orchestrator receives only condensed results, protecting its context from content that would compete for attention later. Thoughtworks (Trial, Radar Vol. 35) notes the evidence is mostly practitioner reports rather than measured token accounting, and recommends context profiling to confirm the orchestrator's context actually shrinks.
Origin
- Coined / introduced by
- Agent builders
- When
- Jun 2025
- Story
- Common in coding agents and multi-agent research systems from 2025; Thoughtworks put 'subagents for better context management' in Trial in Nov 2026.