Office AI advances threaten the rest of employment

Office automation threatens to drag the rest of employment

The promise sounds good: AI agents working faster and cheaper, products dropping in price, and a society spending the same without sweating. The problem is old and remains unresolved. If work ceases to be the mechanism distributing income, someone must invent another. For now, no one has. For its defenders, office job automation is no longer a ten-year projection: it is an ongoing rollout advancing at different speeds by sector, but advancing. The question is what it drags along.

What AI agents already do and what is exaggerated

The starting point is a family of agents capable of performing almost any task currently done by a human at a computer: budgets, invoices, customer service, emails, spreadsheets. The repeated thesis is that the private sector will adopt it "in a hurry" to avoid falling behind, abruptly reducing administrative employment. Some set a date for the starting gun: a new model presented as quasi-AGI, and from 2027, mass layoffs.

The counterpoint is that much of what circulates are demonstrations. A script that succeeds in a three-minute video may fail in a real office, with dirty data and exceptions. It is also recalled that companies publishing these wonders have shares to sell. The response from believers is equally blunt: not everything is demos, and those who doubt are no longer deniers. In between lies an uncomfortable technical detail: when the model truncates information, it does not warn. It continues answering with impeccable confidence. They say it is like working with someone who forgets data without saying which ones.

The tale of the new job and the machine that trains itself

The usual argument —each revolution destroys jobs and creates others— has here, according to its discussants, a scale problem. The steam engine replaced muscle; the tractor multiplied one farmer's output. AI aims to replace judgment, analysis, and basic creativity. It is not a tool needing an operator: it is the operator.

And there is a second tightening. If model training depended on people labeling data for a low wage, that work is already being replaced by reinforcement algorithms and synthetic data in closed loops. AI trains with what AI produces. The hovering question is simple and has no good answer: when we are all users, who feeds the system?

How much GDP can fall and why consumption is key

Here the debate becomes truly economic. The most pessimistic scenario circulating in the thread speaks of a drop of more than 50% of GDP when automation combines with other dynamics. The rebuttal launched is that most economists would agree otherwise: a temporary drop trinc by a product increase due to cost reduction and higher productivity. Both may be true at once.

Because the problem is not how much is produced, but who earns. The chain describing the antiestéticar is mechanical: if a mass of wage earners loses income, they stop buying cars, hiring mechanics, painters, or cleaners. Trades seemingly shielded by being manual are left without clients. A robot producing double buys nothing. It is the dog biting its tail, and the knot is in distribution, not technology.

Guilds, licenses, and surviving trades

From the entire analysis, perhaps the most repeated forecast is that dividing the labor market in two. On one side, trades that can be guilded: professions requiring a license to practice, with colleges and entry barriers. These withstand the blow because their shield is legal, not technical. On the other, those without this umbrella, condemned to precarity and thinning.

The doubt is which side each country falls into. Some antiestéticar institutional protection ends up covering precisely the least qualified posts and exposing those requiring a signature: doctors, judges, technicians. If the Administration protects the easy and releases the difficult, the result is the inverse of what the official narrative promises.

From worker to supervised: the approaching social order

The most uncomfortable scenario posed by part of the thread is not unemployment, but control. If work stops distributing resources, distribution will be done by coercion. Factories, warehouses, and robots are property to be protected from a surplus population, translating into more security, surveillance, and private forces. The worker becomes a supervised subject.

There is literature for this. The Steel Cave, by Isaac Asimov, describes a task not automated precisely to provide employment for people who, otherwise, would be idle. The anecdote is not an ornament: it is the model some see unfolding already. And a repeated warning: it is not just an income problem, it is one of autonomy, with personal data functioning as a tithe.

And what if, instead, it goes well? Some imagine the opposite: more wealth, more leisure, and a basic income for all. The problem is that an idle society forces something for which no one has yet designed a response: what to do with millions of people with nowhere to go at nine in the morning.

---

Voices from the forum

The 'new job' fallacy is a bedtime story for rowers. The history of automation is the history of capital concentration and the destruction of human labor value.

Click to expand...

There will be two sides: the jobs that will be guilded and require a license to be practiced, and therefore survive AI and robotization, and the jobs that will not be guilded and therefore remain decimated and ultra-precarized.

Click to expand...