I will cede this: it indeed seems like you can throw together a shitty half-working prototype very quickly using LLMs. There's been a deluge of half-working prototypes, but years in and where are all these projects now? No study has concluded anything near a 5x improvement.
Besides, my actual opinion is that even if these actually gave a large productivity improvement, it's wouldn't be worth it. The negatives are simply too large:
- the concentration of power they are pushing for;
- the environmental effects of training;
- the disinformation they spread at scale (which in practice is their primary use-case);
- the adverse effect on a shared reality from the fact that they cannot reliably answer questions, yet are presented as doing so;
- the adverse effect on a shared reality from their sycophancy (do the rich surrounded by yes-men seem in touch with reality?);
- their anti-social nature, as they discourage engaging with other people in favour of solipsism masked by a poor simulacrum of personhood;
- their weakening of labour power because of the managerial class' belief that they will soon replace workers (whether or not they actually can is not particularly relevant);
- the large-scale plagiarism they enable;
- lastly, their weakening of already incredibly lax standards of engineering from our field.
The promise of more software does not even come close to outweighing all of that.
I can't speak to you other issues but these are here to stay and I think the only realistic way forward is to learn how to use them.
I guess let's just ignore the evidence, then.
> I can't speak to you other issues but these are here to stay and I think the only realistic way forward is to learn how to use them.
People only say this about things that they want. Ignoring the consequences of one's actions is not a neutral position. Societies are perfectly capable of banning things (yeah, you can probably sneak a local model through a hypothetical law, but it would end the industrial scale problems).
There isn't any evidence. It's pretty much all anecdotal. I can only speak from my experience.
> People only say this about things that they want. Ignoring the consequences of one's actions is not a neutral position. Societies are perfectly capable of banning things (yeah, you can probably sneak a local model through a hypothetical law, but it would end the industrial scale problems).
We're not going to ban these models, though. Which means we have to figure out how best to live with them.
> A recent study of actual coding practices across hundreds of firms finds that human code review forms a significant “bottleneck” for the overall efficiency of AI coding tools, resulting in “little evidence that firms increase software output or reduce employment” by using them.
https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o...
> When developers are allowed to use AI tools, they take 19% longer to complete issues—a significant slowdown that goes against developer beliefs and expert forecasts. This gap between perception and reality is striking: developers expected AI to speed them up by 24%, and even after experiencing the slowdown, they still believed AI had sped them up by 20%.
LLM boosters always cry foul with this one because newer models exist (see next item), but the fact is this is evidence that programmers are remarkably bad at estimating their own productivity.
https://metr.org/blog/2026-02-24-uplift-update/
> For the subset of the original developers who participated in the later study, we now estimate a speedup of -18% with a confidence interval between -38% and +9%. Among newly-recruited developers the estimated speedup is -4%, with a confidence interval between -15% and +9%.
With the rather large caveat:
> Unfortunately, given participant feedback and surveys, we believe that the data from our new experiment gives us an unreliable signal of the current productivity effect of AI tools. The primary reason is that we have observed a significant increase in developers choosing not to participate in the study because they do not wish to work without AI, which likely biases downwards our estimate of AI-assisted speedup. We additionally believe there have been selection effects due to a lower pay rate (we reduced the pay from $150/hr to $50/hr), and that our measurements of time-spent on each task are unreliable for the fraction of developers who use multiple AI agents concurrently.
> We're not going to ban these models, though.
How could you possibly be confident of that?
> How could you possibly be confident of that?
I just don't see any indication that we're anywhere close to that. It would take a rather big shift in governance.