Building agents: When to fold subagents into your main…

Written by AI

TLDR: Anthropic says fewer subagents can improve agent quality, cost, and latency when progressive context and thin tools replace unnecessary orchestration.
Best for: Teams with multi-agent systems that may be adding latency, cost, or inconsistent behavior.
The tip: Keep subagents only where context bloat, parallel work, tool selection, or conflicting instructions measurably justify them.
What you need: An existing agent with multiple subagents
Quality, cost, and latency measurements
Start here: List routers, planners, and domain subagents in the current system.
Test whether progressive context and thin data tools can replace each one.
Keep a subagent only when measurable results justify the separate loop.
Filed note: Anthropic explains when a single agent with skills beats a web of subagents on quality, cost and latency, and when parallel agents still earn their keep.

Highlights: They are effective at verification, parallel work, and modular permissioning.
These are the jobs where subagents remain useful.
In our comparisons across several enterprise deployments, a single agent with skills frequently outperformed both the one-prompt-for-everything design and the subagent design on quality, and often at a lower cost and latency per task.
The simpler architecture won across several deployments.
In these cases feedback is often that most of this structure was doing work the model no longer needed done for it, and that one clear system prompt, 4-5 thin data tools, and firm output guardrails would cover most of the use case.
Many orchestration layers can collapse into prompt, tools, and guardrails.
Excerpt: When building agents in Claude Platform, using fewer subagents in favor of progressively disclosed context and tools can improve quality while reducing cost and latency.
From claude.com