Skip to main content

Envisioning is an emerging technology research institute and advisory.

LinkedInInstagramGitHub

2011 — 2026

research
  • Reports
  • Newsletter
  • Methodology
  • Origins
  • Vocab
services
  • Research Sessions
  • Signals Workspace
  • Bespoke Projects
  • Use Cases
  • Signal Scanfree
  • Readinessfree
impact
  • ANBIMAFuture of Brazilian Capital Markets
  • IEEECharting the Energy Transition
  • Horizon 2045Future of Human and Planetary Security
  • WKOTechnology Scanning for Austria
audiences
  • Innovation
  • Strategy
  • Consultants
  • Foresight
  • Associations
  • Governments
resources
  • Pricing
  • Partners
  • How We Work
  • Data Visualization
  • Multi-Model Method
  • FAQ
  • Security & Privacy
about
  • Manifesto
  • Community
  • Events
  • Support
  • Contact
  • Login
ResearchServicesPricingPartnersAbout
ResearchServicesPricingPartnersAbout
  1. Home
  2. Vocab
  3. Subagent

Subagent

An agent spawned by another agent in its own session, returning a single tool result to the parent.

Back to Vocab

A subagent is an agent spawned by another agent via a tool call, running in its own isolated session with its own context window. The parent agent sends a task and waits for a single tool result in return — not an ongoing collaboration, but a one-shot delegation. The subagent burns through its own context window on the delegated task, reports back the result, and terminates.

Subagents exist to isolate context. When a task threatens to overwhelm the parent agent's remaining context budget — a large grep search, a complex refactoring with many dependencies — spawning a subagent offloads that noise to a separate context. The parent receives only the distilled result, preserving its own attention budget for the primary task.

The critical constraint is that the subagent tree is one level deep. A subagent cannot spawn further subagents, which prevents runaway context fragmentation but also limits the depth of task decomposition. This design choice trades expressive power for predictability.

Open questions include how to handle partial or timeout failures when a subagent task exceeds its context window, and whether hierarchical multi-level agent compositions can be made stable enough for production use without introducing brittle dependency chains.