LangChain Releases DeepAgents — An Opinionated, Out-of-the-Box Agent Harness
Just hours ago, LangChain quietly published a new open-source repository called deepagents — and it's one of the more opinionated things the team has shipped in a while. Unlike LangGraph, which provides the primitives for building agents, deepagents is a fully configured, ready-to-run agent harness that works right out of the box. Install it with pip, and you get built-in planning through a write_todos system, filesystem tools for reading, writing, editing, and searching files, sandboxed shell execution, sub-agent spawning with isolated context windows, and auto-summarization for long conversations. The project is built on top of LangGraph and carries its production DNA.
The release arrives at an interesting moment. Codex, Claude Code, and Claw Code — which just hit 72,000 GitHub stars days after launching — are all competing to become the default runtime layer that developers reach for when they want an agent that can actually do work. LangChain's decision to ship an opinionated, batteries-included harness rather than waiting for third parties to assemble one from LangGraph components suggests the team sees this layer as strategically important, not a detail to be outsourced to the ecosystem.
Whether deepagents gains traction will depend heavily on how it performs against peers that have had more runway — but as a signal of where LangChain sees the competitive landscape heading, the timing speaks clearly enough on its own.