Tony Kim
Oct 27, 2025 05:23
Discover the distinctions between agent frameworks, runtimes, and harnesses in AI growth, specializing in LangChain, LangGraph, and DeepAgents, as detailed by the LangChain Weblog.
The LangChain Weblog not too long ago delved into the intricate world of agent frameworks, runtimes, and harnesses, providing insights into how these elements perform and work together inside AI growth. The dialogue facilities round three distinguished open-source packages maintained by LangChain: LangChain, LangGraph, and DeepAgents.
Agent Frameworks: LangChain
LangChain is primarily recognized as an agent framework, designed to offer abstractions that simplify utility growth. These abstractions function psychological fashions, facilitating simpler onboarding for builders and standardizing the constructing course of. Nonetheless, there are challenges related to abstractions, significantly in the event that they obscure the underlying processes or lack flexibility for superior use circumstances. LangChain 1.0 emphasizes structured content material blocks, agent loops, and middleware to boost flexibility. Different notable agent frameworks embody Vercel’s AI SDK, CrewAI, and OpenAI Brokers SDK.
Agent Runtimes: LangGraph
LangGraph stands out as an agent runtime, important for working brokers in manufacturing environments. It addresses infrastructure-level wants corresponding to sturdy execution, streaming assist, and human-in-the-loop capabilities. Constructed from scratch, LangGraph is designed to be a production-ready runtime, powering agent frameworks like LangChain. Comparable initiatives on this house embody Temporal and Inngest, which give attention to sturdy execution engines.
Agent Harnesses: DeepAgents
DeepAgents represents a more moderen class, termed as an agent harness. It builds upon LangChain, providing options like default prompts, device dealing with, and entry to a filesystem, presenting a extra complete answer than a typical framework. This idea is akin to Claude Code’s strategy with their Claude Agent SDK, though the time period ‘agent harness’ continues to be gaining traction.
Selecting the Proper Device
Understanding the nuances between these elements aids in deciding on the suitable device for particular wants. Whereas LangGraph might be seen as each a runtime and a framework, the distinctions between these classes stay considerably fluid. The event group continues to refine these definitions as they discover new psychological fashions for AI growth.
For a extra detailed exploration of those ideas, go to the LangChain Weblog right here.
Picture supply: Shutterstock
