AI-Generated App Is a Ticking Time Bomb | Hacker News Reader

AI-Generated App Is a Ticking Time Bomb

quantedsquare.com

8 pointsleray_J13 comments

It’s the hype cycle. Are LLMs going to become invaluable to software engineering? Definitely. Are they going to do everything everyone on X says they will? Probably not. Right now we’re in the peak of inflated expectations.

Anecdotally I’ve been using Claude to help me write a C# CLI tool from scratch. The more lines I let it write, the less and less I understand the code. Can I copy/paste it and it works? Probably 90% of the time. When I have to go and fix it, it is a huge burden.

When I prompt it to do one singular function, it’s amazing. That’s a clear and concise unit to understand.

This matches my experience. There's a place for LLMs in exploratory research. There's a place for LLMs in scaffolding code that would otherwise be too tedious to write.

But where real decisions need to be made by those with "skin in the game", LLMs are a disaster. Vibe coding quickly falls off a cliff. They create code architectures that no human would given the context, and as a result no human can feasibly understand or maintain them long term.

I would say this is boon to real software developers - tons of apps that need an expert to maintain, it's job security, right? Well, no. I suspect, rather than maintain these LLM-generated monstrosities, most organizations will just let them rot. That was the default in the golden age of zero-interest engineering, that's the default in the golden age of vibe coding too.

When it's too rotted, organizations plan to replace the whole system and pay 10x the price of what it would have cost to do it right in the first place, costly migration consulting mission, bullshit stuff and noise everywhere. The mid/long-term game is where the opportunity lies if you are a dev. The cycle is kind of safe for software engineer to me. Jus there won't be a place for HTML/CSS integrator, untalented PHP dev copy / pasting of random PHP function from SO anymore as it use to be 15 years ago. Those are cooked.

There are several steps missing from:

A startup founder uses AI to create an MVP

They secure funding based on the demo

They find themselves unable to move beyond that initial prototype

I have trouble believing tech investors throwing money towards a founder that shows a vibe coded product without anything else. The product might be vibe coded, BUT the founder show some traction or discovered a new market or something besides a demo. It does not really matter how a founder did the MVP, they did not create google. They just showed, through the MVP, something to investors. What happens next, after money, i doubt it's vibe coded.

I agree with your point about the funding process being oversimplified. However, I've seen many founders become emotionally attached to their MVP, convinced it's good enough to scale without fundamental rethinking. They push forward with what's essentially a throwaway prototype instead of rebuilding it properly with lessons learned from initial user feedback. And some people are just cheap by nature. Last point is a huge one, even when funding is good enough. Does not make sense to make, I've seen it.

>However, I've seen many founders become emotionally attached to their MVP, convinced it's good enough to scale without fundamental rethinking.

i have no doubt, but this has nothing to do with the MVP being vibe coded or not - this has something to do with the (technical)experience of the founder + other personality traits

I see pre seed round going to millions before there is an app. And I've been called as last resort to help people fix their app in the weekend because national TV was coming next Monday.

I can 100% see AI worsen and accelerate that kind of things.

Exactly my point, this has nothing to do with how the MVP was made. Dubious investment rounds exist with or without vibe coding anyway.

I would have agreed a few weeks ago. But I can't anymore, the limitation I see is scope size and not capability. And available context sizes will just get bigger

Have you seen LLM play chess? They would not best an average 6 years old who just learned the rules. I doubt giving them more context would help. There are some thing you can't just parrot your way out.

And I would easily argue that it's harder to copy the behavior of a good software engineer than it is to copy the moves of a good chess players.

Not even talking about understanding the business logic you are actually implementing. Which LLM seems as far as understanding chess.

As I see it I am directing it while constantly watching, controlling and questioning it.

I wouldnt let it do anything on larger code bases, but building highly specialized services or components IS crazy efficient, when you question ever move.

Well, until Claude Code ends the game(maybe in a few months), yeah. Today we're not there yet, even with full codebase knowledge. At least from my experience and how I want it to be done. For some, the damage has already been done. Wait & see !

I've seen it do so stupid things I am still afraid running each query. It honestly kinda stressful to keep up with, when it starts breaking things. More like a super fast trainee than a coworker to depend on. Yet do damn efficient to work with

Reply on news.ycombinator.com