AI will create more jobs than it destroys, debate concludes

The thesis being defended: AI multiplies the startups that end up hiring programmers

The example used to illustrate it: an electrician who runs his own business, handy around the house, who always wanted to learn to code and who, thanks to AI, sets out to build his own system. When the project gets tangled and he can no longer handle it, he has put 30,000 euros of his savings in and has two options left: hire programmers or lose the money. That is the mechanism described to explain why artificial intelligence, instead of emptying the tech job market, would be manufacturing demand for people who know how to fix what it wrote itself.

Why does AI generate jobs instead of eliminating them?

The central argument is that AI replicates human cognitive abilities but has no causal reasoning. It can program, yes. But it is inconsistent as the project grows: the more complex the system, the more blunders it makes and the less efficient it becomes. The result, according to this reading, is a business that starts without developers and, once it reaches a certain size, needs them.

On that basis, a specific founder profile emerges: people with money and no technical background. The guy who buys apartments and rents out rooms, the self-employed handyman, the nurse who discovers she can build her own SaaS. They all get so far on their own. Then they call an engineer.

And they don't push on out of calculation, but out of pride. Someone who has told their family they're going to be the next Bill Gates doesn't shut down the operation — they hire. As the person holding the thesis sums it up, lies and deception, even self-deception, are more powerful than money.

Code nobody understands: the maintenance problem

The objection is harsh. Against that thesis, another participant argues that code written by an AI is not understood by a human, because it is not born from a mind that gets tired, forgets details and needs simple, immediate logic. An uncomfortable consequence is pointed out: no engineer can get you out of the mess when you depend on a mountain of code generated by another mountain of code.

The image circulating is that of Doctor Frankenstein: someone has to know which corpse each leg came from, that is, which instruction each module came from. Then you need the lightning bolt, and that's where luck comes in. A similar diagnosis admits that AI already programs faster than a human, but not better: what it doesn't do is think better.

How much does it cost to set up a local AI?

Against dependence on the cloud, the alternative proposed is running models at home. The problem is the price. The RTX 5090 launched with an official price of 2,300 euros and is currently selling for between 4,000 and 7,000. A Mac Pro with 128 GB of memory is on the same path. And with that you can only run a few experiments: local models are more quantized and slower compared to the cloud.

The relief, if it comes, would be RAM production bringing prices down. Meanwhile, some argue that the entire supply is being snapped up by cloud AI at outrageous prices, and others sum it up with a classic: the cartels are at it again.

The scenario that throws everything off

Against all of the above, some bet on the machine: if it already solves problems in quantum physics, fluid dynamics and genetics, it will eventually program infinitely better than a human. That is the optimistic bet, the one that holds the bottleneck will disappear on its own.

At the other extreme, another participant sums it up without nuance: if the work of 8 is now done by 1, "forget about it." The balance, in that reading, would be negative.