The One Announcing Paradise and the One Announcing the End Are Selling the Same Thing

In February, during the Super Bowl, Anthropic spent millions of dollars on a run of ads mocking its competitor for putting advertising inside its chatbot. The message was clear: Hours later, OpenAI launched the advertising anyway.

Thirty seconds on that broadcast costs more than eight million dollars. So the two most important artificial intelligence companies on the planet used the most expensive television event in the world to argue with each other about advertising, by spending on advertising.

I'm not telling this as industry gossip, because the gossip is already old. I'm telling it because that scene explains what's happening quite a bit better than any of the prophecies you'll read this week about AI advancing into our lives.

Twelve months

"Within twelve months, the most intelligent being on the planet will no longer be a human."

Mo Gawdat, former Google executive, said that a few days ago, on a stage.

That same week I listened to Yuval Noah Harari, the Sapiens guy, explaining something more specific and considerably more dystopian. He says AI won't take power with killer robots à la Skynet, but through bureaucracy. The systems that actually govern the world are banks, registries, markets, legal systems. Machines made of language: laws, deeds, accounting entries. We never liked them and we were never good at running them (anyone who hates Excel is right there), but we ran them anyway because there was nobody else on the planet who could hold up the scaffolding.

We're bad at finance, he says, but horses are worse.

And out of that comes the most "relevant" part of his talk. A horse's life today is governed by the financial system, and the horse doesn't even know that system exists. It looks around and sees trees, dogs, houses. The market is invisible to it because it's too complex for its head. It lives inside something that determines it and that it will never understand.

At first it sounds trivial. Then it has a twist that lands closer than it seems: the new horse could be some of us. The ones who hit accept on everything the machine proposes, without reading, fourteen times a day.

And it isn't a new idea, in case it feels familiar. Éric Sadin calls it vegetalization: delegating so much cognitive agency that you end up a passive node inside a system you don't understand. One arrived by way of the historian, the other by way of the philosopher, and they both end up standing in the same place.

I agree with that diagnosis. With the specific case used as proof, I don't.

The data point almost nobody checked

In May, the Argentine government sent Congress a draft of a new General Companies Law. It's not a partial reform: it repeals the law that's been in force for nearly fifty years and replaces it entirely. Among other things it creates a new figure, the Automated Company: a company that can carry out its purpose through autonomous algorithmic systems or AI agents, without needing human intervention for ordinary operation. It carries the word "Automated" in its name, has full legal personality and limited liability, and answers with its own assets for damages its systems cause.

Harari publicly came out to warn about this. He described it as Argentina being about to grant legal personhood to artificial intelligence, with direct access to the financial and political systems.

The bill doesn't say that. But it sounds like a well-staged horror film.

The legal personality belongs to the company, not to the AI. The shareholders are still natural or legal persons. The text does not recognize an agent as a subject of rights. Daniel Vítolo, one of the jurists who has worked the topic most, clarified that under current legislation those companies still need administrators who are people. And the bill is in Senate committees, so all of this can still change.

I don't point this out to play fact-check police. I point it out because the error runs in exactly the direction that makes the thesis more dramatic. A company operated by algorithms, whose owners are still human and which answers with assets, is an interesting piece of news. An artificial intelligence recognized as a person before the law is a much better piece of news for filling a room.

Follow the trail

Let's go back to the Super Bowl, and look at the numbers instead of the intentions.

Tech companies spent 333 million dollars on television advertising about artificial intelligence in the United States during 2025. Up 43% from the previous year, and this year that hasn't changed. OpenAI ran a sixty-second spot for the second year in a row. Anthropic spent some sixteen million on television alone.

At the same time, OpenAI projects losing around fourteen billion dollars this year across compute, research and infrastructure, and has an IPO on the horizon with a valuation being discussed in twelve-digit figures.

Put those two things together: a company burning fourteen billion a year, heading to the stock market, that needs the entire world to believe what it built is the most important thing that has happened in decades.

You don't need to assume a conspiracy. It's enough to look at who needs the story to be big.

And on the side of those issuing warnings, the same thing happens through a different mechanism. The bigger the problem you describe, the more important the room where you describe it. Davos doesn't invite you to explain that adoption gets stuck in the security committee. It invites you if you come to say the end of human work is upon us and how that might be profitable for "someone".

The one announcing paradise and the one announcing the apocalypse are, without having agreed on it, selling the same product: that this is enormous and you need to look now.

And meanwhile, downstairs

Here's the part of all this that makes me laugh.

In any large company you've ever set foot in, this will sound familiar. A few years ago I spent three months waiting for a twenty-dollar license to be approved. Not twenty thousand. Twenty. Three months of forms, mail chains, people asking for the business justification and other people waiting for the first group to answer. And it wasn't the exception: that's how the machine works, at that company and at almost all of them.

Now line up the altitudes on these 3 points:

Gawdat says twelve months until the machine is smarter than any of us.
Harari says that afterwards it keeps the systems.
The enterprise software vendor promotes something called, literally, Autonomous Enterprise, with two hundred-odd agents ready to execute business processes.

And at the very bottom there are a few of us waiting three months for a signature on a twenty-dollar license. Or for permission to deploy an agent in production.

I'm not telling this to be a smartass or to belittle those narratives. I'm telling it because that distance is the data point. I've worked for more than twenty years inside one of those bureaucratic machines. I don't study it: I operate it. And in all those years I never once saw a system take itself.

What I always saw was the same thing: somebody authorizes.

Bureaucracy isn't seized. It's opened.

In Harari's description he starts with cave-dweller societies and ends with an omnipotent AI taking over humanity through bureaucracy. Something "better" at administering (than us) shows up and keeps our system, like a fitter species displacing another. It's an evolutionary image, which works beautifully on a stage, and it operates on your subconscious without you questioning it.

But inside, it doesn't happen that way.

An agent doesn't get into a production system because it's smarter than whoever operates it. It gets in because somebody gave it credentials. Somebody decided what it can call and what it can't. Somebody defined what gets logged. And that somebody has a name, an employee number, and a boss who will ask them about it when something goes wrong.

Bureaucracy doesn't fall by conquest. It opens by authorization. And every authorization has a signature underneath it.

That changes the scenario completely, and it stops being what is AI going to do with our systems, and becomes something far more boring: who is signing, with what judgment, and what happens to that person if something breaks.

An agent can execute, can decide, can optimize. It can't answer for it. And the whole edifice that worries Harari is built on the idea that there's always someone who answers.

Before you agree with me

Three things that cut against what I just said, and I'm leaving them here because otherwise someone else will raise them with less care.

Bureaucracy defending itself is not a virtue. The three months for twenty dollars also block good things. The same slowness protecting you today from a badly configured agent is the one leaving you behind a smaller competitor. You can't ask for both.
The human signature can become a formality. Someone hitting approve fourteen times a day without reading, because their inbox is full. There the name is there, the formal responsibility is there, and the decision is nowhere. That happens today, with no artificial intelligence involved.
I am not neutral. I work in technology, I live off this, and I push AI adoption inside my own area. When I say human judgment is still the bottleneck, I'm describing a world where my own trade still has value. Keep that in mind when you read me.

What is worth doing

And here I want to close differently from how I've been closing, because if I leave you with nothing but distrust it's no use to you.

None of this is an argument against using artificial intelligence. I use it every day and it makes me better at several things. It's one of the most powerful tools we've been handed, and staying out of it from fear or badly calibrated skepticism is as bad as buying the whole story.

The point is different, and it's more boring than the two versions being sold from the stages: adapt without handing over your judgment.

Use it to think better, not to stop thinking. Argue with it. Ask it to explain why it chose that path. Keep the reasoning, not just the result. And above all, watch the reflex of hitting accept for the quick reward, which is what these tools do so well you don't even notice.

Harari's horse didn't lose because something smarter showed up. It lost because nobody asked it anything. Because the system governing its life was built without it in the room, and by the time it existed, there was no way in.

We're still in the room. We sign, we authorize, we grant credentials, we decide what gets exposed. It's not a comfortable or glorious position: it's boring work, about permissions and lists.

But it's the room. And the day the signature becomes a formality somebody hits without reading, we won't have been replaced. We'll have walked out on our own.

Pablo / En Producción

P.S. If you work somewhere that someone is going to have to sign off on this, send it to someone else who'll also have to sign it. There are fewer of us than it looks.