Building your first applications with LLMs is daunting because you need to interface with many data sources and build prompt workflows. So you’re wondering, “should I start with a development framework like LangChain?” For most teams, the answer is … no. Not now.
These products are supported by great engineers and top-tier venture funds. But unfortunately they are just too early for most development teams. Building an abstraction layer on a new technology stack is extremely hard because there are no standards. The foundational models like GPT-4 are changing constantly, and there are no agreed-upon best practices for building solutions. It will take years for these standards to emerge, and until then the abstraction layers like LangChain will be in constant flux.
If you don’t believe me, spend some time on Reddit and you’ll see some strong opinions.
What should you do? Here are my suggestions:
Ultimately, every team soon realizes that building your own interfaces isn’t that hard and ultimately elects to do so.