This is true, but also importantly this only works if you carefully redefine productivity to mean something else than a craftsman would consider productivity. You need a simple metric that's easily cheated.
For example, a software engineer who cares about what he does would define productivity fuzzily, as general growth of functionality for the consumer of the application, with the implied "actual working well-crafted functionality". If you're an idiot who wants to hack productivity, you define it as something straightforward and stupid, e.g. lines of code added. Suddenly you can claim that an "AI software engineer" is more productive than a human.
This exists even in something seemingly all about quality, such as research. One of the many problems with the current state of academia is the obsession with "number of papers published" to the disregard of rigor, and so you'll get people who are more interested in hacking the metric than actual research. Hence the seemingly annual scandal where someone is caught completely fabricating data, or the even more frequent sham experiments in psychology that never replicate. The replication crisis falls into the same category -- it's good science to replicate, but journals are not interested so it doesn't grow the sacred metric by which every academician is judged.
Unfortunately we're in an age of hacked productivity. The productivity metric for our markets is line going up, which has long been disconnected from actual productivity, as in providing a product to customers that willingly buy it. It's hard to keep focus on actual productivity when seemingly everyone around you, and especially everyone hierarchichally above you, cares only about the hacked metrics. Art is one of the few mainstays where you alone can be the judge of your own productivity and whether you're happy with your output, since at the core the only metric that matters in art is "does it feel right to me". This must be untenable to promptfondlers because they never experienced actual artistic fulfillment, so instead they need a hacked metric to feel good about improving -- how many images can we churn? how long of a video can Sora output before killing itself? how many seconds of "music" can our box generate?
I never thought I'd say this but... don't slander category theory like that, compared to LLMs it's downright useful