I’ve talked a lot about AI in hypotheticals, although it is certainly having real-world effects in various fields already. Nevertheless, nothing else quite hits the same as it occurring at your place of business.

For context, general staff where I work have been using ChatGPT for more than a year now. Mostly, it has been used to summarize their own notes, check grammar, and so on. All of which had been expressly against policy, considering how the notes in question are filled with sensitive personal information that has now been consumed by OpenAI servers and possibly regurgitated into someone else’s results. There’s no fighting the tide though, so the executives worked with various regulatory agencies to bring in a “walled-off” version of (I think) Microsoft’s version of AI into the fold.

Things took a particularly different track the week before the holidays.

See, there have been a number of initiatives and requests over the years from staff to get electronic versions of paper forms, and/or to modernize the electronic forms they already use. Our IT shop is small for an organization of our size, with just 2-3 programmers on staff for such requests. The latest form to be modernized has taken almost two years to be completed, for… various reasons I won’t go into here. So when a new request came in for a different form, it was not much of a surprise to learn that the estimate was ~2,000 programming hours. For us, that sounded about right.

And then… the senior programmer just completed the entire form in about 100 hours via AI.

During the impromptu “demo,” he outline his methodology, which was essentially guiding the AI into building each specific section of the form separately. Interestingly, as each section was built and tweaked by the programmer, the entire codebase was (apparently) re-entered into the prompt to ensure later sections integrated with the earlier pieces. All of which took the stated 100 hours and around $50 worth of AI credits. I think this was via Github Copilot, on one of the lower “settings.”

Brilliant, right? If this outcome is reproducible with other forms – and the form in question doesn’t self-destruct down the road – it will effectively save the organization millions of dollars over time. There is nothing but unambiguous good about it, yeah? People dream about a 20x improvement in efficiency!

So, here’s the thing. The form we’re talking about already exists as a PDF and Word document. It now also existing in what amounts to a company website could possibly save some amount of staff time, but the overall return on investment by any metric was dubious from the get-go. Is going from 2000 to 100 impressive? Yes. Would I expect our junior programmer to be able to reproduce the same 100-hour outcome? No. Is there a very real possibility that the next round of hirings will exclude junior programmer positions? Yes. Does anyone have grand designs for how junior programmers will become senior programmers in the future? No. Will future AI use for a similar effort also cost $50? Fuck no.

If you’re pessimistic about an AI future, you’re a Luddite, fighting an absurd battle to keep buggy whips relevant. “Get with the times, old man! The future is now!” Sure, okay. Question, though: do you feel we are well positioned, politically or economically, for the fruits of AI investments to improve the lives of the working class? Or perhaps are we just automating the creation of shit sandwiches?

Alas.