Very interesting how people view disruption differently.
Personally I am very much against IP (have been since 2017), and think decompilation and recompilation projects being accelerated is pretty awesome.
With no frontier company able to attain a monopoly on model quality (they keep leapfrogging eachother), I think there is a mitigated (albeit not totally eliminated) risk of centralization, which would be the biggest concern of this. Open models are pretty good and can be accessed quite cheaply.
I understand why this might bother people who fall more on the stability and order end of the "order vs. chaos" spectrum though.
Personally, I see AI coding (as somebody who writes code on a daily basis) as no different from any movement up an abstraction layer. Most people coding in the last two decades aren't touching assembly code, as an example. Yes, some functional knowledge is lost to most in that. But it is fine, because other knowledge and technique is gained. They still can produce quality high-level code, in so much as they know how the algorithms and data structures they're handling work with a minimal level of detail. What happens with going even another level up to natural language is that the work of "coding" is moved to the systems engineering level, and the focus is on systems design. Those who are good at that still have a technical advantage.
As an example, in the last year especially but to a degree even before then, much of my day to day has been building out semantic-layer data models, standardizing metrics, etc. These are things AI agents still struggle with to an extent, because they often have blurry internal world models and can't sift through business context as seamlessly as a group of humans can. A lot less of my time is spent thinking about a specific module, API, or any other syntax quirk. Which is fine. That was the more frustrating part of the job anyway. Eventually (maybe a year from now; maybe a month) coding and design agents will be good at these sort of metric definition and semantic modeling tasks. Then we move up another layer of how to clearly communicate problems and needs.
And yes, probably some day in the next decade or two, most white collar work will be automated or automatable. The necessary implication of that, if we allow our social institutions to evolve with the technology, is a realistic version of the FALGSC meme. The key is that we have to evolve socially, and I am sure that is also scary to the people who are more order-motivated and stability-sensitive.