Someone at work was telling me the other day about how they had attended a demo of some LLM-powered tool that “will write documentation just from looking at all the code.” They went on to suggest that we could point this tool “at all the old legacy stuff that we don’t have people to support.”
“Why would we do that,” I asked. “Are we going to hire people to go read all that documentation?”
The answer, of course, was no.
So if we don’t have any documentation for any of this code now because there are no people to support it, having some LLM thing that automagically creates a bunch of docs does nothing for us. We would be better off spending whatever money that costs on just shutting down this old stuff instead of piling more technology and tooling on top of the problem.
And I feel like that is really the issue that all of this LLM crap comes down to. In the course of doing my job, most of the times I hear people talking about “Could we use AI for that,” whatever “that” is really just a candidate for “How about we don’t do that anymore?” I don’t need Copilot to summarize slide decks for me; I need fewer slide decks. Conversely, I don’t need Copilot to generate slide decks for me; I need fewer people asking me for slide decks.