LangChain's new deepagents file format cuts edit errors 15% and tokens 10%

LangChain · x · 2026-09-15

LangChain's OSS team changed how deepagents reads files, reporting that the new format reduces editfile errors by 15% and total input token usage by 10% based on their evals.

The team frames this as part of an ongoing effort on "small but mighty" agent harness optimizations, arguing these incremental improvements compound significantly in agent workflows.

Original post →

More from coding & agent

coding & agent channel →