DEV Community

Discussion on: Is "Vibe Coding" Ruining My CS Degree?

Collapse
 
bbkr profile image
Paweł bbkr Pabian

"Team vibe" is straight way to unemployment.

In a few years it is not you who will be vibe coding. Your Product Owner / Product Designer will. They can use keyboard too to write prompt with feature request. You will be an obsolete layer in this process.

The demand for developers will be reduced to hardcore stuff. Backbone systems, mainframes, very complex backends, device drivers, bare metal code, etc. You will be paid to understand complex problems.

Collapse
 
maame-codes profile image
Maame Afua A. P. Fordjour

You’re absolutely right

Collapse
 
darkwiiplayer profile image
𒎏Wii 🏳️‍⚧️

This is assuming the current model is sustainable. With the AI bubble continuing to grow yet nobody finding a way to make money from it, while the internet is getting saturated with already AI generated content and the people generating the quality training data are getting fed up with their work being stolen and starting to take measures against it, I wouldn't be too surprised if AI crashed really hard in the coming years and only made a full comeback after many years of hardware breakthroughs make it financially viable again.

Collapse
 
bbkr profile image
Paweł bbkr Pabian • Edited

It is sustainable. Or to be more precise - it reached a level which makes it reasonably priced tool for software development.

AI bubble will crash hard. All AI FOMO and AI Slop projects will be evaluated by the market, which is natural healing process of oversaturated economy. But AI development tools will survive this crash for three reasons:

  • They are way less expensive to train than generic LLM models.
  • They have good ROI. They speed boring things up, like summarizing legacy code, making documentation, suggesting next lines, rapid prototyping, etc.
  • Compilers are great, instant feedback loop for AI models, guarding against whole range of errors and allowing to correct mistakes quickly.
Thread Thread
 
darkwiiplayer profile image
𒎏Wii 🏳️‍⚧️

They are way less expensive to train than generic LLM models.

This faces the same feedback problem as most other AI training: Real people start moving their code out of reach from AI tools and more and more of what remains is itself AI-generated. Eventually the quality just degrades.

They speed boring things up, like summarizing legacy code, making documentation, suggesting next lines, rapid prototyping, etc.

And in the process make mistakes more likely.

Compilers are great, instant feedback loop for AI models, guarding against whole range of errors and allowing to correct mistakes quickly.

Compilers don't catch difficult mistakes. If your standard for good AI code is that it compiles, then yea, compilers make that easy; but for one they aren't "instant", specially on bigger projects, and more importantly, they don't help much with the real problem of AI code.


But all of that aside, my point still stands. AI doesn't pay for itself; it's not financially sustainable. And that's likely going to get worse as the massive resource consumption brings AI into conflict with local populations around server farms.

The bubble will burst eventually, AI will "die" for a while, and eventually, technology will have caught up to its requirements and companies will start to scale up its usage again, probably with much less hype surrounding it.