alt Hacker News> I still do not see how these copy-machines with "super intelligence" inside are going to replace humans
I do.
You don't need any sort of fancy harness as claimed.
All you need to do is type vague ideas into a text box. Then you need some UX taste to pick good concepts.
The building of the software happens trivially now. You do not need a software engineer, sorry guys.
What you need next is someone with time and patience to QA test the software and provide feedback for the AI.
I guess ideally the same person, with the UX taste.
Give it a few weeks and the product is done, without the need for a team of 5-10. All it takes is one competent human.
I know that many here vehemently disagree and raise concerns about the viability and maintainability of the software.
I believe not a single one of them has made that experience, these concerns are based on their imagination of what they think would happen. Usually they tell you that they essentially review every line of code in their projects.
Replies
The problem with this is that it is not written by a human. It’s one of the versions of: „Since AI I finished all my spare time projects.“ Etc.
The actual reality is, that it is too early to actually say where this is going to end. Either AI is going to fix what it is causing right now or not.
We are screwed either way. So is the person or prompter behind the message above.
If Ai is universal there is no need for UX, because a real personal assistant, that takes personal care of you is the only UX necessary in the end.
>The building of the software happens trivially now. You do not need a software engineer, sorry guys.
This part is quite "load-bearing".
I heard from someone who worked at Lovable that what most customers produce with that magical textbox is software that shouldn't go anywhere near production. Apparently the problem of thinking like an engineer is not solved by these AI tools. Most people who are not software engineers by trade are not capable of verifying what they produced, nor would they even consider needing to.
No offence, I think you might be a classic case of "senior who forgot how it feels to be junior", where after a lot work experience deva start taking their software engineering instincts for granted and forget that they are a skill that takes years to master. You can prompt an AI for a solution, you can even prompt it for ideas, but you cannot prompt it to think for you.
context is important. Depending on what you are doing, you might have to review every line of code two or three or ten times before you sign off on a PR.
> All you need to do is type vague ideas into a text box.
If all you give the machine is a vague idea, then it is unlikely that you will get what you want or need. Pure information theory. Garbage in, garbage out. You could get lucky, but luck based strategies are just frustatingly bad in the long run.
> The building of the software happens trivially now.
The building of bad software does, no doubt. I'd like to use good software and my sincere hope is that AI tech will be used to raise software quality substantially since most of what I see around me and have to use every day is just really bad, compared to what it could be if people cared.
> You do not need a software engineer, sorry guys.
For quality you do.
I think this lands in the same area as flying jet liners. A computer can do it. But legally it seems easier if there's a human responsible for the operation of the aircraft. I've conjured my ideas into code via LLM, and it is magical. But the quality is also off, and there are nontrivial issues that I'd like to address. The more a software project is vibed, the less one can manually address things.
Often, the agent will fix a problem without assessing if the fix is worth it, optimal or defective. It's fine to merge fixes like this in codebases occasionally - humans have been managing to do this and to work around this for ages. I'd like to see a software project that is purely LLM driven and doesn't devolve into chaos because of this.