[gyomu]

You can build software fast, but the more time you take the more you can be confident about it being correct.

This was true already before AI code generation, but AI code generation means the definition of “fast” has shifted like crazy. An experienced engineer can now build in an hour what would have taken them a week.

But it (right now) does nothing for the confidence about being correct - you still have to put in the extra time to ensure the software is correct to your desired degree of confidence.

Unfortunately for people who like their software to be correct (stable, performant, bug free, etc), it seems like way too many software developers right now are happy to take the gains of the former but ignore the tradeoffs of the latter.

(at the same time software quality for mainstream consumer goods like the kind of thing OP complains about in their post has been down the drain for a long time now, and it seems likely many of the examples they give are pre-AI)

[bodge5000]

> An experienced engineer can now build in an hour what would have taken them a week.

From what I've seen, I don't think so if we're talking about like-for-like quality. Maybe as an absolute best case scenario, but more typically it's more like you can build in an hour with AI whereas without AI it might take you 2 or 3.

Even thats highly task dependant, and doesn't scale as you'd expect (eg 1 hour == 2 hours, but 1 month =/= 2 months), and at times even using AI can slow things down, at its worst case more severely than at it'd speed you up at it's best.

Now if you're willing to throw like-for-like out the window, that 1h==1w could be achieved easily, maybe even more. But then again, I feel like I could do a lot more in an hour if I didn't care about quality either, not as fast as AI but much faster than I am currently.

[foldshift]

I think I can say I got some 10x speedup with the recent, more powerful models (from Opus 4.5 on). But then again, I am just building CRUD websites, although sometimes there are some challenging tasks like some fancy graphs/plots to show, and some API integration in the backend, etc... but these are the use cases that models are really good at. I have 0 worries about scaling: my apps are internal, so you can think of 10-15 concurrent users at most, usually 5 or less, and 100 or so users in total, which eliminates the whole class of complicated scaling problems. 99% of the devs are probably on the same boat: our day-to-day problems are not complicated, dont need any deep algorithm or data structures knowledge, and models speed up this type of work a lot. Of course, they get very limited on complicated problems

[hollowturtle]

Totally agree. But we're in such minority that it's super depressing

[bodge5000]

I like to think, or hope, that its just a vocal minority and that the majority of engineers are at least more considerate with their AI use than going all in. It kind of makes sense too, "I programmed X entirely using AI" is a more compelling proposition for an article than "I programmed X the same way people have been doing for decades", so of course you're going to hear more from them.

[vouaobrasil]

> From what I've seen, I don't think so if we're talking about like-for-like quality. Maybe as an absolute best case scenario, but more typically it's more like you can build in an hour with AI whereas without AI it might take you 2 or 3.

The real problem is that the current paradigm in software design is to aim for the bottom of the barrel or a minimum working thing that can make some money and then move onto the next thing. I think in a way AI is a direct outgrowth of that mentality, because so little emphasis is placed on the final 5% of polish that brings software to the level of amazing.

It's much more evident in photography, which is what I started doing after I quit software. You can see that people are editing photos faster and better with AI, at least at first glance. But the end result is almost always sloppy compared to someone who takes time to think about their edit. I think that basically characterizes AI: yes, it makes things faster (or so I'm told, I've never used an LLM), but it also SHIFTS the emphasis onto rapid prototyping of superficial things.

Can those things be polished? Maybe. But it doesn't change the fact that less and less polishing will be done and people will focus more on creating new, superificial things in the long-term and we'll lose a lot of potentially great, hand-crafted things. Not to mention that AI adoption will likely accelerate even more levels of frameworks, which in turn means more hardware upgrades.

[teaearlgraycold]

Really depends on the engineer, domain and circumstances. I find the best multiplier from AI when working with tech I’m familiar with, in a specialty I’m less familiar with, on a brand new project. I have the vocabulary to describe what I want and how to do it but the details I’d need to spend hours learning are handled for me.

There are absolutely projects I’d done in a week a few years ago that could be done in 3 hours now. But if you need to work with that code for the foreseeable future you’ll need to actually read the generated code.

[lelanthran]

> You can build software fast, but the more time you take the more you can be confident about it being correct.

> ...

> But it (right now) does nothing for the confidence about being correct - you still have to put in the extra time to ensure the software is correct to your desired degree of confidence.

You're partially correct - what gives confidence is duration in the field. No matter how long you take to build something, seeing it work in the field for 3 months gives me more confidence.

You have to release, early and often.

[hollowturtle]

And break ux for users often and piss them. There are short reversibile decision that can be taken fast, important decision requires a lot of thinking before changing to often and alienating your users