The AI crash is impossible - Change my View

[Natasha Tori Maru]

@bazera Interesting analysis.

Do you think with AI replacing low level jobs, new entry level people entering the workforce are missing out on learning from the ground up to be able to troubleshoot, asses and review what AI produces? An experience gap is being created.

Or do you think the system will need a reshuffle to train and teach undergrads to be able to assess and review what AI produces?

It appears a gap could be created where there are those with the wisdom to know when there is code slop, and those who do not have the experience to know what is shit and what isn't. This would necessitate a different set of tools being taught prior to entering the workforce.

Dunno, just spitballing. I am not a coder at all.

Edited 2 hours ago by Natasha Tori Maru

[Leo Gura]

AI has not replaced a single serious coder.

[UnbornTao]

8 minutes ago, Natasha Tori Maru said:

Dunno, just spitballing. I am not a coder at all.

Anyone can be now!

"Code a coffee ordering app now, make no mistakes."

Edited 2 hours ago by UnbornTao

[Natasha Tori Maru]

5 minutes ago, UnbornTao said:

Anyone can be now!

"Create a coffee ordering app now, make no mistakes."

LEGIT lol

It just seems to me like anything in life, the experience of learning from the ground is essential to mastery. AI doing graduates jobs might create a gap in skills; if we shortcut we usually miss wisdom. In this case, the ability to critically assess code.

Edited 2 hours ago by Natasha Tori Maru

[Joshe]

2 minutes ago, Leo Gura said:

That is not how complex software can be made.

Someone just created a Figma clone in a weekend, supposedly. If you're not familiar with Figma, it's a highly complex UI design tool. I doubt it's production ready, but even to get a working prototype of Figma in a weekend is already crazy. If that can be done in a weekend, just imagine what you can do in 1-3 or even 6 months.

It's all about systems. The more the AI has to think about every minute detail (clutters its context), the more it will err. So you have to solve structural/foundational problems early and lay building blocks/patterns to separate concerns. For example, build all your user interface components first so the AI doesn't have to process frontend and backend at the same time. Then, impose architectural rules for how things get wired up. Then, have a QA process in place to check for errors + a pre-mortem. There's a lot more to it than this but I hope you get the point.

You wouldn't have the AI build a UI component on the fly - the UI component should already exist so CC can simply place it where it needs to go without much thought, allowing it to focus on architecture, state, and wiring stuff up. It's on the developer to be able to spot the situations in which CC might err, which only comes with experience. Being able to read it and track what it's doing gives you a massive edge. 

"Vibe coding" fails because it's approach mixes concerns every step of the way. But if you have a full system in place and the AI is clear on all the rules, and the implementation is intelligently tested and coordinated, you most certainly can build complex software in this way, and it is being done. 

One thing you might not be aware of is you can have multiple agents running in parallel, all working on different features via git worktrees. This requires strategic planning, but it's how the Figma clone was built in a weekend. It takes intelligence and strategy to be able to orchestrate this, but this is now possible. 

I can't speak to how good it would be for gaming projects, but here's a video I saw the other day from a game developer who said it made him 10x more productive, and he has the receipts. I think you'd change your stance about AI coding agents if you actually dove deep into it. 

 

[Leo Gura]

12 minutes ago, Joshe said:

Someone just created a Figma clone in a weekend, supposedly.

Supposedly is the key word.

People are making slop. Not real software. Yes, you can create slop apps that copy existing apps poorly. This has no value.

Edited 2 hours ago by Leo Gura

[UnbornTao]

25 minutes ago, Natasha Tori Maru said:

LEGIT lol

It just seems to me like anything in life, the experience of learning from the ground is essential to mastery. AI doing graduates jobs might create a gap in skills; if we shortcut we usually miss wisdom. In this case, the ability to critically assess code.

Yep. The allure of a convenient path is apparently too hard to resist.

What you're describing seems to be happening, to some degree, with Windows PCs now. I'm not even sure Microsoft has beta testers for the OS - or even that they have many developers and coders these days. They just wing it and treat users like guinea pigs.

Edited 2 hours ago by UnbornTao

[bazera]

17 minutes ago, Natasha Tori Maru said:

Do you think with AI replacing low level jobs, new entry level people entering the workforce are missing out on learning from the ground up to be able to troubleshoot, asses and review what AI produces? An experience gap is being created.

It's not that AI literally replaced en emplyee, it just reduced a need for more junior roles at certain companies. 

Also, the expectation and requirenments for an entree level coding job increased tenfold over the last couple years. Plus Covid made many people realize that coding jobs was profitable and everybody starting learning it, like, in my city if you asked random people if they were learning how to code, 7 out of 10 would probably say yes. And the market got oversaturated pretty quickly. Plus many people was laid-off after Covid was settled and that also created a huge competition on the market.

So there were lots of factors in the fact that now juniors struggle to find a job. It's not really only because of AI advent. But it plays some role.

To answer your question, yes definitely, that's the real danger that I see today, many people are missing out on learning from the ground. 

Just imagine trying to learn math and you copy all the answers from the answers sheet for 80% of the times for the problems, because nobody taught you that it was irresponsible, you just figured that it was easy. 

I remember when I was starting my career, at job, I was having nightmares over the issues I had left at job because there was no way of solving them other than my own mind and research capabilities which I had to develop through painful experiences over time. But now you can just ask a chatbot to "please fix", and you're done. 

That's something that really concerns me. It takes lots of responsibility to use these tools correctly, and you know the level of responsibility that an average person has, so yeah.

Quote

Or do you think the system will need a reshuffle to train and teach undergrads to be able to assess and review what AI produces?

There needs to be some of that as well, but for 90% of the times, undergrads should use their mind to solve problems, not AI. Hell, even experienced people need to do that. Sometimes I feel like my mind is rotting when I use LLM's for coding, it's still a new problem so I'm still trying to figure out what's the correct way of using them for productivity and also not to rot my mind.

Quote

It appears a gap could be created where there are those with the wisdom to know when there is code slop, and those who do not have the experience to know what is shit and what isn't. This would necessitate a different set of tools being taught prior to entering the workforce.

That gap is already here. Not with just coding, also with other things like AI videos, music, etc.

For example, when I listen to some AI generated music, it may sound good to my ear and ignore it's AI, but when professional musician will listen to it, that will be much different. Same with the code.

For example, yesterday I was tasked to modify some of the feature I worked on last year. I did it with Claude Code in 15 minutes, when without it I'd need at least 4 hours to do it. But it was done with couple iterations, first I asked it to give me a plan, then I modified some things in it's plan, then the first implementation was a bit sloppy (which I was able to identify through existing experience), then on second and third iteration, it was something decent to be pushed and released. 

I had to meticulously review every line it generated, or else it might halucinate something, and if you aren't careful, customers will complain and if that happens many times, you might even lose customers.

So business will lose customers if you aren't being careful as a developer. And you won't be able to be careful if you don't have experience. 

[bazera]

20 minutes ago, Joshe said:

"Vibe coding" fails because it's approach mixes concerns every step of the way. But if you have a full system in place and the AI is clear on all the rules, and the implementation is intelligently tested and coordinated, you most certainly can build complex software in this way, and it is being done. 

No it's not. 

No software that actually has customers is being made / maintained solely by AI, that's just not the case.

Maybe if you have a small startup with a small app with few features, yes. But it still needs an experienced person to overlook it.

Edited 2 hours ago by bazera

[bazera]

It's all fun and games until your "fix it please I beg you" prompts doesn't work anymore and you have to debug and fix manually the code you didn't come up with yourself. Lol.

I've been there. All the AI fantasies crash at that point.

[Natasha Tori Maru]

@bazera troubleshooting and creative problem solving are the consistent elements through my career that have meant I have excelled in everything I have really applied myself to. And resilience. Really driving at a problem. Sticking with it.

So your words resonate across huge domains.

19 minutes ago, UnbornTao said:

Yep. The allure of a convenient path is apparently too hard to resist.

What you're describing seems to be happening, to some degree, with Windows PCs now. I'm not even sure Microsoft has beta testers for the OS - or even that they have many developers and coders these days. They just wing it and treat users like guinea pigs.

I feel like this is happening to Outlook as a microcosm of your point about Windows OS. I had to go into my registry to deactivate the new version and revert to old. I have no idea how it made it to live in the form it is.

[Joshe]

18 minutes ago, bazera said:

No it's not. 

No software that actually has customers is being made / maintained solely by AI, that's just not the case.

Maybe if you have a small startup with a small app with few features, yes. But it still needs an experienced person to overlook it.

Of course you need a human to manage things. My entire comment made that clear. 

But there are many people with little technical knowledge making money from AI-coded solutions right now. Just because you aren't doing it doesn't mean it's not happening.

The video I posted shows a 10-year SWE transitioning from "not letting AI write any of my code" to "AI writes most of my code and has 10x'd my output". Watch that video and let me know what you think. 

Edited 1 hour ago by Joshe

[UnbornTao]

3 minutes ago, Natasha Tori Maru said:

@bazera troubleshooting and creative problem solving are the consistent elements through my career that have meant I have excelled in everything I have really applied myself to. And resilience. Really driving at a problem. Sticking with it.

So your words resonate across huge domains.

I feel like this is happening to Outlook as a microcosm of your point about Windows OS. I had to go into my registry to deactivate the new version and revert to old. I have no idea how it made it to live in the form it is.

God, the new web app. 

[Natasha Tori Maru]

Just now, UnbornTao said:

God, the new web app. 

Rake me over hot coals and pour acid onto the wounds. I'd rather that. 🤮

[UnbornTao]

3 minutes ago, Natasha Tori Maru said:

Rake me over hot coals and pour acid onto the wounds. I'd rather that. 🤮

Have you heard of our lord and savior Linus Torvalds? 

[bazera]

@Joshe Sorry if I misinterpreted.

I'm subscribed to his channel and watched that one when he posted it, he makes good points.

Quote

AI writes most of my code and has 10x'd my output

This is exactly my experience as well.

Quote

But there are many people with little technical knowledge making money from AI-coded solutions right now. 

That might be the case, but that doesn't mean it will be sustainable long term. Also, "making money" isn't a good indicator to me for quality software.

Slop can make money as well.

The project that I'm working on has been in production for 8 years as of now. It takes a really careful usage of AI to not mess it up over time. 

It won't be apparent that you're messing up the project with 1-2 promtps. It will be a slow process.

If you let AI coding assistance run amok only with people with little technical knowledge on project like that, it will become disaster in couple months.

Maybe it can work for smaller scale projects.

Edited 1 hour ago by bazera

[bazera]

16 minutes ago, Natasha Tori Maru said:

troubleshooting and creative problem solving are the consistent elements through my career that have meant I have excelled in everything I have really applied myself to. And resilience. Really driving at a problem. Sticking with it.

Me too. AI could really boost that process, it could assist you in improving your solutions, give you new perspectives, help you when you're really stuck and just wasting time with no hope.

The key word in "AI assistant" is assistant. I'm trying to use it as assistant, not just for solutions generator.

So it's not really an AI issue, it's more like a people issue.

People are irresponsible, and giving irresponsible people these tools without proper guidance might go sideways.

Edited 1 hour ago by bazera

[Joshe]

29 minutes ago, bazera said:

The project that I'm working on has been in production for 8 years as of now. It takes a really careful usage of AI to not mess it up over time. 

All good! 100%. You'd have to be very careful in that scenario.