The people who write AI slop would have written human slop, they're just much faster now, and the people who are really good tend to be so set in their ways that they resist AI when they could quadruple their productivity if they just put half as much time into learning how to use it as they put into their current skillset.

Great engineers can quadruple their productivity with LMMs? Strong claim.

If their whole job is throwing together disposable demos and 1-use scripts, I could believe 4x. But in the normal case, where the senior engineer's time is mostly spent wrangling a legacy codebase into submission? I just don't see LLMs having that level of effect.

Depends on how coupled the codebase is. Heavily coupled code is hard for LLMs to modify, but brownfield work that's not heavily coupled, you can easily refactor and improve coverage at a much higher rate with LLMs.

You are omitting a third class: the people who would not have written anything at all but are now writing AI slop because the bar has been lowered.

Pretty much the reason I think electronic music was fantastic right up to the point DAWs became a thing. When everyone with a laptop and some software licenses can just put out slop that sounds like the originators, quality fell off the cliff. 90s and 2010s electronic music are a world apart.

Yeah, it seems to be a problem, especially when it comes to 'discoverability'. Just like the app store: I hate browsing app stores like Apple/Google Play and stopped doing it entirely because every query has 40+ different apps doing very similar things.

You're not wrong but there's entire jams on YouTube from people with real gear and no DAW that produce uninteresting synth based music.

Yet The Prodigy made good albums entirely with Reason.

I'm not sure what point you're making. I never claimed any set of tooling is necessary of sufficient to produce good music. I merely pointed out that, as it became more accessible to produce music, the overall quality of the output decreased.

I think the volume of good music has actually increased, but maybe less than the bad music.

Honestly there are good reasons to resist AI even as a senior. If prompting an LLM cannot be easily integrated into your existing workflow or demands a huge change in tooling, it's just distracting or crippling. I don't think we need to have a long discussion as to why distractions and context switches are counterproductive.

You can say that, but when the principle down the hall from you who's got a highly tuned agentic AI workflow is outproducing half your team by himself while leading engineering direction on multiple projects, it's hard to justify that to management.

it's an "if", not a "when"

this outcome is far from certain (unless you're a slopper, in which case it's obviously happening tomorrow)

I’m sure it will all come down to their risk appetite.

The AI slop will continue as long as the stock market rewards it.