At Octomind, we use AI agents with multiple LLMs to automatically create and fix end-to-end tests in Playwright. Until just a few months ago, we did it with the LangChain framework.‍In this post I’ll share our struggles with LangChain and why replacing its rigid high-level abstractions with modular building blocks simplified our code base and made our team happier and more productive. The backstoryWe used LangChain in production for over 12 months, starting in early 2023 then removing it in 2024...