Skip to content

Sub-Agent Spawning UPDATED

Problem

Large multi-file tasks blow out the main agent's context window and reasoning budget.

Solution

Let the main agent spawn focused sub-agents, each with its own fresh context, to work in parallel on shardable subtasks. Aggregate their results when done.

Recent developments show that improved agent state externalization capabilities may make subagent delegation more practical by helping agents better identify which tasks are suitable for delegation and how to communicate necessary context to subagents.

Example (YAML front-matter refactor)

sequenceDiagram MainAgent->>GlobTool: "*.md" MainAgent->>TaskTool: spawn 4 sub-agents with 9 files each loop per SubAgent SubAgent->>Files: update front-matter SubAgent->>Git: commit end MainAgent->>Git: merge branches ➜ single PR

Advanced usage: Swarm migrations

For massive parallelization (10+ subagents), see the Swarm Migration Pattern which extends this concept for large-scale code migrations.

High-volume use case at Anthropic:

Users spending $1000+/month on Claude Code are typically running swarm migrations:

  • Main agent creates comprehensive todo list
  • Spawns 10+ parallel subagents
  • Each handles batch of migration targets (e.g., 10 files)
  • Common for framework migrations, lint rule rollouts, API updates
  • Achieves 10x+ speedup vs. sequential execution

Quote from Boris Cherny (Anthropic):

"There's an increasing number of people internally at Anthropic using a lot of credits every month. Spending over a thousand bucks. The common use case is code migration... The main agent makes a big to-do list for everything and map reduces over a bunch of subagents. You instruct Claude like start 10 agents and then just go 10 at a time and just migrate all the stuff over."

References

Source