@arrasz Basically its a sequel to the software rot we are already seeing: The project has a budget, it has a deadline, we need to ship it, so corners are cut, training is skipped and refactorings only cost money. I know that's not how you are approaching software, and its not how I approach my own projects, but it does happen in the industry very often. And I don't see those companies going out of business because of that.
@javahippie
The slightly more complicated question is what it will do to to developers over the next couple of years.
Developing software is so much more than writing code.
So if I, as an individual develper or we, as a company decide not to use it due to ethical reasons, we *will* be at a disadvantage on the job market and the software market. Lets not lie to ourselves, I'm not putting out artisanal code that is beautiful to look at, ultra-adaptable and will scale indefinitely. That's rarely the requirement, and doing so would already overstep our clients budgets.
@javahippie I think it puts "don't write maintainable code, write disposable code" in a new perspective.
@javahippie This raises a very important point: If you decide not to use it due to ethical reasons, you will be at a disadvantage. I'd argue that no company acts ethical for ethics sake, at least not for very long. That leaves us with regulation.
@oli I agree. Regulation will fail to handle open models you can self host, but that shouldn't be an argument against it, that's the problem with regulation.
Which regulation do you have in mind? Training on copyrighted material, using LLM for sensitive topics?
@javahippie All of the above I'd say. To put it bluntly: If your model is trained without explicit consent from the data owner you're dealing stolen goods and we have laws for that. Same goes for any form of decision: There should be no way to pass responsibility to the model or its provider.
But since we obviously can't even pass policy to prevent child labor I'm not that optimistic.
@javahippie I wouldn’t judge to harshly before testing offline running or fairly designed models that undercut ethical reasons.
There is a lit coming up that enables you to use the good stuff without the bad stuff.
@dalcacer That's a possibility, but requires hardware investments and will increase your own electricity bill. Also, just because a model is "open", doesn't mean its not trained on intellectual properties of others.
@javahippie I would argue that the hardware/co-processors are part of next gen PVs anyway.
Also there are initiatives like https://ethz.ch/en/news-and-events/eth-news/news/2025/07/a-language-model-built-for-the-public-good.html
@javahippie having played with those tools a lot I totally second your thoughts. And imho not using them is a big mistake. Using them „correctly“ and „responsible“ is a whole different discussion though. I can imagine that some customers will even pay premiums for software deliberately NOT written using them. Time will tell.
@myfear It's like the hand-formed artisanal 48h sourdough bread for 500g at 7€. I also think there will be software that is so critical that needs full attention and a lot of skill. The Salesforce integration into the dispo system that calls a REST Service and writes data from system a into system b? Not one of those, that needs to work and it needs to be finished, soon.
@javahippie Coming to the conclusion, too.
However you forget deskilling and lock-in.
I can happy write all the software I imagine in VI, even a Spring Boot app. Nobody does, but it's fine and works. I know so many devs that are not as lucky as my using IntelliJ under JavaChampions license but pay for it… This is a lock-in too.
Wait until all the vendors of AI assists crank up the prices for good. It will be slaughter.
@rotnroll666 @javahippie Deskilling in particular will destroy the junior-to-senior pipeline. Unless companies invest in it, which is the opposite of what we see happening right now.
Companies are taking a loan up front they will end up paying in maintenance and outages.
AI can be powerful in the hands of a skilled dev. But they will retire, quit or get laid off.
@rotnroll666 For individual projects? Agree completely. For teams at work? Raise the monthly subscription to 1000$, if you simply look at the output, would still be worth it.
I think the Open Source models will become important in the future. That's a heavy investment in server infrastructure, but will seriously make business hard for the SaaS providers
@javahippie @rotnroll666 wouldn't bet on this. The hardware requirements are quite enormous, add maintenance, fault tolerance etc. And you'll be paying for hardware that will usually be barely utilized. At least for smaller to medium companies I think some *-a-service variant makes more sense. But I also don't see much of a lock-in. If one gets too expensive, switch. The competition is massive and the differences already are diminishing and will only become smaller. The same is true for OS.
@javahippie @rotnroll666 define “output”.
That said, 99% of business software is run-of-the-mill CRUD stuff. Having humans writing the same stuff over and over does sound wasteful.
@larsrosenquist @rotnroll666 "Created features per day", as sad as it sounds.
@javahippie @rotnroll666 Well, it’s not as sad as lines-of-code. But only barely. 😂😱
@javahippie Different area, deskilling via React and friends, kinda reverse: People forgot / never learned any basics of JavaScript and HTML, and know only this.
@rotnroll666 This will be a big, big issue in the future. But since when are companies known for planning for the future? They need revenue now
@arrasz As is your right 😉 I'd be interested why? I think that my hand written code still beats the one generated by LLMs, I'm just wondering if that is needed for every application. My bread and butter are enterprise integrations, this stuff has not changed in 10 years, knows its scale upfront pretty often... I'm not sure how I can justify me taking 20 times longer with added value
@arrasz Basically its a sequel to the software rot we are already seeing: The project has a budget, it has a deadline, we need to ship it, so corners are cut, training is skipped and refactorings only cost money. I know that's not how you are approaching software, and its not how I approach my own projects, but it does happen in the industry very often. And I don't see those companies going out of business because of that.
Phew, i would say then i had tons of Lück with my customers the last 15 years.
Another point is that pricing and dependencies will lead to a time where a Junior is way cheaper than an AI coding thingie. Thoughts?
@arrasz My suspicion is growing that there are two entirely different software industries, and I'm in the bad one :D
@arrasz Hard to predict. For that to happen, the subscription needs to be 2000€/month +, I don't think that's viable. But you can put together a server with some graphics card for 15k and host the LLM there, the "open" models are quite good already.
I'm pretty sure that we will see an education crisis, or already are seeing it. The students in my Masters module last year were relying on ChatGPT for a lot of things heavily, they just use it naturally, like I used Google back then
@javahippie a colleague said,its like an apprentice who drinks every night. 😎🍻
@arrasz I think its like a very seasoned senior who drinks. For me it produced mid-level code, but very complete features I could have not written by hand in a day in seconds. But sometimes it produces code that tries and sends private keys to a different server 😅
@javahippie readable Code it is my friend. Write once, maintain years!
I haven't seen thus from an AI thingie
