Writing software completely from scratch is inefficient. That’s why it’s so important that documentation includes plenty of example code. Developers seeking to onboard new users to their fancy new framework have more or less figured this out.
But things are starting to change. Without having to browse documentation or Stack Overflow first, I can instead ask a free chatbot to write some code for me in a language I don’t fully understand and then iterate as needed, undercutting the need to dive into documentation as deeply or frequently as I would otherwise need to. Does this mean documentation is outdated?
No.
In fact, documentation will probably become even more important as it starts to serve not only humans, but LLMs too. In a few short years, we may see developers specifically advertising how their language, npm package, or API works well with GPT-X. Artificial intelligence will provide a basic scaffold for a software project, but it will still be the responsibility of the humble human developer to write the complex bits. And whether both parties are capable of fulfilling their role will still require thorough documentation to do so. In theory, this should actually encourage developers to describe their software very explicitly to help the AI understand its functionality, in much the same way people try to use SEO techniques to make their food blogs more visible to search engines.
I hope it doesn’t happen, but I wouldn’t be surprised if OpenAI sold a “GPT-ready” certification to companies trying to woo developers by showing how fine-tuned and calibrated GPT is when handling their software.