LangChain has re-architected its package into multiple packages: langchain-core, langchain-community, and langchain, to improve developer experience. The langchain-core package contains simple core abstractions and the LangChain Expression Language, which allows users to compose these components together and expose streaming, batch, and async interfaces. This is done in a backwards compatible way. The langchain-community package contains third-party integrations, while the langchain package will focus on higher-level and use-case specific chains, agents, and retrieval algorithms. LangChain also introduces the concept of runnables, which are sequences that expose the same interface with single, batch, streaming, and async methods. This design is useful because it allows users to write a runnable once and invoke it in many different ways. The re-architecture aims to address issues such as complexity, stability, and versioning. LangChain also highlights its ecosystem of templates, LangServe, and LangSmith, which provide tools for building and deploying GenAI applications.