Source record

Published by LangChain Blog (T1.5 source). The original is at https://www.langchain.com/blog/organizing-context-in-a-multi-agent-harness.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

LangChain's deepagents introduces forked subagents that inherit the supervisor's full conversation context, instead of starting with an isolated context window. Forking reduces redundant context-gathering, leverages prompt caching, and cuts cost and latency. Isolated subagents remain the default when context isolation is preferred.

Concrete architectural choice for multi-agent systems: when to fork supervisor context versus isolate. Tradeoff is redundant work against context pollution, with caching and cost implications.

Cited by

No citations on record.