why we no longer use LangChain for building our AI agents

Octomind, a company utilizing AI agents with multiple LLMs to automate end-to-end tests, initially used LangChain for its framework but encountered difficulties due to its rigid high-level abstractions, which complicated code comprehension and maintenance as their requirements evolved. The team found LangChain's abstractions, while initially helpful, soon became a hindrance by adding complexity without tangible benefits, and they struggled with its inflexibility in a rapidly changing AI field. This led them to abandon LangChain in favor of a more modular approach, using simple, low-level code and selective external packages, which allowed for greater adaptability, faster development, and reduced friction. They concluded that building AI applications without a framework, focusing on core components like LLM communication and observability platforms, is a more effective strategy. This approach enhances their ability to innovate and iterate, aligning better with the dynamic nature of AI development.