LangChain Explains Three Frameworks in Three Minutes; Commenters Say Picking One Takes Three Weeks

The official LangChain account posted a three-minute video whose title is just three names: LangChain, LangGraph, and Deep Agents. The video aims to answer a very specific question: what exactly is the relationship between these three things?

The comments are more informative than the video. Dominykas Bickus wrote: three minutes to explain, three weeks to choose. Guy said they keep confusing LangGraph and Deep Agents. Hershal Rao said that after watching these three minutes, two new frameworks might have come out. Michael Kaiser said finally someone explained the difference. A few replies make the state of the ecosystem clear: explaining is explaining, choosing is choosing.

The three names are easy to confuse because they are not on the same level. LangChain is the entry point: integrations for models, tools, and retrieval are connected here first, making it suitable for quickly building a working LLM application. LangGraph manages state and flow, and is used in multi-step, multi-agent scenarios that need interruption and recovery. Deep Agents is higher up, built on top of LangGraph, and aimed at deep agent scenarios that require capabilities such as planning, sub-agents, and a file system. The names sound like three products, but they are actually three layers.

Ordinary users don't need to memorize all of them. If you want to connect models quickly, start with LangChain. If you want to make the flow clear, look at LangGraph. If you want agents to break down tasks themselves, then look at Deep Agents. Practitioners should care more about the layering than about taking sides. Task complexity, team maintenance cost, and whether to lock into the ecosystem are more important than names.

The official video can clearly explain what they are, but it cannot choose for you. That comment about three weeks is not a joke; it is the real cost. The more lively the framework ecosystem and the more names there are, the heavier the cognitive burden. Hershal's complaint makes sense: you've just memorized one set, and the next set arrives.

So this video is worth watching, especially for newcomers. It puts the three names back into one picture. But what three minutes solves is what they are, not which one you should use. For ordinary users, first know what problem you want to solve, then remember the names. For developers, first manage the flow and state well, then talk about deep agents.