LangChain's own docs show an interface rebuilt since its 2022 start

The conversation

LangChain began as an open-source library in October 2022; its own repository on GitHub records that origin, and describes the current project as 'a framework for building agents and LLM-powered applications' that lets a developer 'chain together interoperable components and third-party integrations.' The project has since grown into a versioned ecosystem, and its documentation at docs.langchain.com, retrieved 16 September 2026, now centers on a function called create_agent rather than the chain abstraction the project first shipped with.

What the documents show

The README states LangChain's purpose as providing 'a standard interface for models, embeddings, vector stores, and more,' so a team can swap a model provider with limited code changes. The current overview describes create_agent as 'a minimal, highly configurable agent harness' composed from a model, tools, a prompt and middleware, and states that LangChain's agents 'are built on top of LangGraph,' a separate, lower-level orchestration library the same organization maintains for durable execution and human-in-the-loop workflows. Read together, the two documents show an interface that has been restructured, not merely extended, since the project's 2022 debut.

The system boundary

LangChain's documentation positions the library as a layer above a model provider's own API: it defines a standard interface for calling a model, storing embeddings or invoking a tool, and the underlying execution still happens inside the model provider's endpoint or the tool's own code. LangChain's role, as its own documentation describes it, is orchestration and portability across providers, not a capability the underlying model does not otherwise have.

Where it fails

Because LangChain's own documentation confirms its abstractions have been revised across major versions, code and tutorials written against an earlier chain-based interface are not guaranteed to run unmodified against the current create_agent and LangGraph-based design. Neither cited document states a backward-compatibility guarantee across major versions; a team standardizing on the library is accepting some exposure to further interface changes as the project continues to evolve.

• Which LangChain version and abstraction, chain-based or create_agent-based, does a given piece of code actually target?
• What does switching model providers through LangChain's standard interface change about latency, cost or tool support?
• Where does LangChain's orchestration end and the underlying model provider's own execution boundary begin?

LangChain's repository and current documentation establish what the library provides today; neither claims that provision has been stable in shape since October 2022.