There's of course been a race to solve customer support and documentation with AI. OpenAI's own docs are decidedly non-AI:

And even the king of AI frameworks, LangChain, outsources it to a third party:

There's Kapa.ai, Kyber, Forethought/SupportGPT, Scrol.ai, and more. If you want open source DIY solutions, there's DocsGPT, and any number of Langchain and Llamaindex tutorials.

And, as far as I can tell, that is it for the docs/support AI crew. However, these approaches have some downsides:

• they are evidently much slower than traditional search, and yet not a lot better than search

• costs are borne by the host company

• most of the bots are run based on automated docs scrapes, which can fall out of sync

• the amount of prompt engineering and extra context you can add to a third party bot is quite limited and likely doesn't fully support your usecase

How Docs must change with AI

Most people in the docs world know about the Divio 2x2:

But I've lately been inspired by Metaphor Systems' approach, which adds a third dimension:

Here it is in the docs:

and how it looks when used:

Docs GPTs bring a number of benefits:

• Here the traditional divisions between Tutorial vs Reference vs Explanation vs How-to's get smashed. The user doesn't have to find what they want, they can just ask their question.

• GPTs can code, especially if you can bring-your-own-interpreter in future, so working code for your exact usecase can be produced from docs GPTs.

• The user runs the GPT on their own OpenAI account, so it doesnt cost the host anything. User also brings information on their own personal chat preferences which immediately makes it a better UX

This is exciting and after trying out the Metaphor GPTs for a while I am wishing that every company offered their docs this way.