Through many years of working with APIs, I’ve always been struck by the fundamental problems of SDKs. Most are not well designed or built. Many are not complete. Few represent actual use cases or intended workflows. And yet all can introduce supply chain risks into our apps. A decade ago I tried to address this through “SPOIL your Users with Great SDKs” but it wasn’t enough.
With generative AI, I saw a new option in our toolbox..
How could we get a Large Language Model (LLM) to do the work and generate a fully featured SDK for us? Would an OpenAPI spec be useful? Going further, is it feasible to build unit and integration tests? How about instead of trying to define what “good” means for our particular SDK, we just found existing good examples and referenced those? And for bonus points, could we have our systems do code review of the generated code to refine and improve it over time?
With that in mind, I recently launched a course “Building Better SDKs with Generative AI” available exclusively from LinkedIn Learning.
In this course, I start with the OpenAPI spec from GitHub and build a Python SDK by:
- starting with a simple prompt
- directing it to use an established coding standard
- generating unit tests
- refining the code based on security and performance suggestions
- leveraging other LLMs for collaborative code review
- and more.
In only ~90 minutes, I lay the foundations, give you the vocabulary, and get you functional with SDK generation. Check it out and let me know what you think.
Note: I started this reasoning in early 2024 with How ChatGPT will solve all API problems.. except yours and continued it later in the year with How to Kill your SDKs in 1 Easy Step.