LangChain Deep Agents: A Batteries-Included Agent Harness for Complex Tasks
LangChain has released Deep Agents, a production-ready agent harness built on LangGraph that comes equipped with planning tools, filesystem backends, and subagent spawning capabilities for handling complex, multi-step tasks.
Unlike traditional agent frameworks that require developers to wire up prompts, tools, and context management from scratch, Deep Agents provides an opinionated, ready-to-run agent out of the box. The harness introduces three key innovations: a planning tool that lets agents decompose tasks and track progress, a filesystem backend for persistent context management, and the ability to spawn subagents with isolated context for independent subtasks.
LangChain coined the term "agent harness" to distinguish this from lower-level frameworks and runtimes. Where LangChain is the framework and LangGraph is the runtime, Deep Agents is the harness — a batteries-included layer that handles orchestration decisions so developers can focus on their application logic.
Early benchmarks are promising. The deepagents-cli (LangChain's coding agent) improved 13.7 points on Terminal Bench 2.0 (from 52.8 to 66.5) by only tweaking the harness while keeping the model fixed, demonstrating that harness engineering can deliver performance gains without model upgrades.
GitHub: https://github.com/langchain-ai/deepagents | Docs: https://docs.langchain.com/oss/python/deepagents/overview
← Back to all articles
Unlike traditional agent frameworks that require developers to wire up prompts, tools, and context management from scratch, Deep Agents provides an opinionated, ready-to-run agent out of the box. The harness introduces three key innovations: a planning tool that lets agents decompose tasks and track progress, a filesystem backend for persistent context management, and the ability to spawn subagents with isolated context for independent subtasks.
LangChain coined the term "agent harness" to distinguish this from lower-level frameworks and runtimes. Where LangChain is the framework and LangGraph is the runtime, Deep Agents is the harness — a batteries-included layer that handles orchestration decisions so developers can focus on their application logic.
Early benchmarks are promising. The deepagents-cli (LangChain's coding agent) improved 13.7 points on Terminal Bench 2.0 (from 52.8 to 66.5) by only tweaking the harness while keeping the model fixed, demonstrating that harness engineering can deliver performance gains without model upgrades.
GitHub: https://github.com/langchain-ai/deepagents | Docs: https://docs.langchain.com/oss/python/deepagents/overview