LangChain is older than ChatGPT. Harrison Chase created the repository on October 17, 2022, as a side project while working at the ML startup Robust Intelligence - six weeks before ChatGPT launched and turned “connect an LLM to tools and data” into everyone’s problem. The timing made it the default: when the whole industry suddenly needed prompt templates, retrievers, and tool calling, there was already a library with a name for each of them.
The numbers since then are unusual for an open-source library. In spring 2023 the project became a company, with a seed round from Benchmark and a Series A led by Sequoia. In October 2025 LangChain raised $125M at a $1.25B valuation, led by IVP, and reported 90 million combined monthly downloads of langchain and langgraph and usage at 35% of the Fortune 500. Today the Python repository has more than 146,000 GitHub stars. And it was never Python-only: LangChain.js has existed since early 2023, which matters if your product lives in Node.js and TypeScript.
It also became the most criticized framework in AI engineering. Early LangChain wrapped everything - prompts, chains, agents, memory - in layers of abstraction, and many teams found themselves debugging the framework instead of their product. Some publicly ripped it out. The team listened: LangChain 1.0, released on October 22, 2025, was rewritten to be smaller and more opinionated. The core is now one entry point - createAgent in TypeScript, create_agent in Python - running on the LangGraph runtime, with middleware hooks around model and tool calls, and a promise of no breaking changes until 2.0.
The ecosystem is really three products. LangChain is the high-level API for models, tools, and agents. LangGraph is the lower-level runtime: agents as graphs with explicit state, durable execution, and human-in-the-loop steps - that is where long-running workflows live. LangSmith is the commercial part: tracing, evaluations, and deployment, and it works even if you do not use LangChain at all. In practice LangSmith is often what teams keep after they simplify everything else.
My rule of thumb: for a single model call with structured output, I do not need a framework - the OpenAI or Anthropic SDK plus a Zod or Pydantic schema is shorter and easier to debug. I reach for the LangChain ecosystem when the problem is genuinely agent-shaped: several tools, retries, branching, state that must survive a restart, a human approving a step before money moves. That is where LangGraph’s explicit graph and checkpoints save more code than they add, in TypeScript and Python alike.
A detail I like: the thing LangChain sells today is not chains at all. The name comes from 2022, when “pipe one prompt into another” was the frontier. In 2026 the product is agent engineering - runtime, observability, evaluation. That is the real lesson for anyone choosing an AI stack: pick tools for the problem you have this year, keep the model call behind your own interface, and make sure you can trace every step when something goes wrong.
Give the next product someone who stays.
Ready to get started? Send me a message - I'm available to start now and to stay. Long-term retainers, not a two-week handoff. Let's discuss your project, requirements, and the best way to move forward.