The comparison with cooking was not casual: in Italy a chef in an average restaurant (not a Michelin-star level one) can have a reasonable wage and live in relative confort, while a McDonald's guy would struggle.
Now, people don't need a restaurant to have a healthy diet, but they need decent ingredients and a bit of cooking culture (something most of people breath since early age in Italy... except the poorest people who can't afford that).
Yet, to get new working software you need programmers (or to become one), because these #LLM can only output lossy decompression of their source data, #opensource software and all the proprietary source code that their users submit¹.
So I'm pretty confident that #junkcode will remain a quite apt definition of #vibecoding output for the forthcoming decades².
Meanwhile those who "leverage" these new tools will slowly loose their ability to understand what their code does, while those who don't will retain and increase such ability, becoming relatively more valuable as programmers.
The problem in the short term is the spread of junk code in #freesoftware, as it will inevitably lower its quality (security, maintainability, and so forth).
In the middle term, we will find new ways to protect the commons that will necessarily violate some old tabu.
Years ago I wrote the #HackingLicense to this aim:
The Purpose of the Hacking License is to create a cultural and technological corpus that will grow and evolve completely and exclusively as a common heritage of humanity._____
¹ I find amusing how corporations that allow the use of #ClaudeCode or #Copilot do not realize how they are paying big money to work for #Anthropic, #Microsoft and so on.
It's overly naive (and ignorant) to believe their model are improving over time because of technical breakthrough or increased computing power.
The percieved quality of these models depends on how well their output meld into your proprietary codebase, and it's quite obvious that the more of such codebase feeds their training data, the more you will percieve that the model capabilities have grown.
That's why such companies operate at loss: your money help pay the bill, but your work, your code, your reviews, improve their product!
And the funny thing is that when your company pays for your #Claude subscription, all the code of your company will be compressed and distributed to all of your direct competitors!
And they will be all like "look how good is the code from Claude!"
In practice Anthropic&friends are exfiltrating and distributing your code to your competitor and you pay them to do so!
🤣🤣🤣
Sure, you get access to your competitor code too, but guess who win this game?
Hint, neither you nor your competitors! 😉
² I mean, until the #AIbubble pop. If you remember the effects of the subprime crisis in 2008, you might have an idea of what we are going to face, except that it will be an order of magnitude worse, in paricular in the #USA.
One might hope that such depression will make #US people get rid of their pluticracy in a similar way French got rid of their aristocracy in late 1700.
But unfortunately US people already live in a #fascist state, and they are just learning how twisted is their illusion of freedom
