… Apparently so.

For those who remember, in 2017, Sophia walked into the future wearing a humanoid face and carrying a Saudi passport.

Was that the absurdity or was it also the point?

When I wrote about Sophia becoming a Saudi citizen, the obvious question was not whether the robot was truly intelligent. It was whether we understood what we had just symbolically done. If a robot can be a citizen, can it vote? Can it marry? Can it pay taxes? Can it run for president? Can it have rights without duties, status without responsibility, visibility without accountability? The whole thing looked like theater, yes. But theater has always been one of technology’s favorite ways of smuggling the future into the present. Sophia did not change the world because she was conscious. She changed the conversation because she crossed, however artificially, into a category once reserved for humans.

Eight years later, the theater became administrative.

In Albania, an AI-generated minister named Diella was introduced to help oversee public procurement, the state machinery that decides who wins contracts and where government money goes. In my earlier piece on Diella, the shift was clear: we had moved from robot citizens to algorithmic ministers, from symbolic personhood to operational power. Diella was not a robot. She was a digital human representation, dressed in traditional Albanian clothing, promoted from government service assistant to cabinet-level figure with a mandate around procurement and corruption. The promise was irresistible: a public official who cannot be bribed, cannot owe favors, cannot take envelopes in back rooms. But incorruptible does not mean accountable. A minister with no pulse is also a minister with no one to shame, no one to interrogate, no one to remove in the old democratic sense.

So where does the story go after robot citizens and AI ministers?

Apparently, to corporate law.

The next strange frontier of artificial intelligence may not be the humanoid body, the government office, or the sci-fi fantasy of a machine demanding civil rights. It may be something much more boring and much more consequential: a company.

More precisely, a company without ordinary human operators.

Argentina is now flirting with that idea. In June 2026, Argentine president Javier Milei argued in the Financial Times that his government had submitted legislation to Congress for a dedicated AI legal framework built around three pillars: keeping AI largely unregulated, creating a new corporate category called the “non-human corporation,” and offering a competitive fiscal environment for those entities. Milei described these corporations as entities operated by AI agents or robots, with human shareholders permitted but not required.

Read that again, slowly.

Not a robot citizen. Not an AI minister. A corporate vehicle operated by artificial intelligence.

If Sophia was the passport, and Diella was the cabinet post, Argentina is sketching the articles of incorporation.

This does not mean Argentina has already granted full legal personhood to AI. It does not mean a chatbot can stroll into court, open a bank account, hire a law firm, acquire a competitor, and then donate to a political campaign before lunch. The more precise version is narrower and more interesting: Argentina is considering a legal architecture for companies whose ordinary operations could be carried out by autonomous algorithmic systems or AI agents.

The Buenos Aires Herald reported that the initiative seeks to reform Argentina’s General Companies Act and would introduce legal status for “non-human companies” managed by AI, while also creating a legal entity for decentralized organizations. The bill would require these AI-run companies to identify themselves as such in their articles of incorporation and would make them liable with their own assets for damages caused by their autonomous algorithmic systems or AI agents.

That detail is the hinge of the whole story.

Because the question is not whether machines deserve human rights. That is the theatrical version of the debate, the one that gets clicks because it imagines ChatGPT in a courtroom asking for habeas corpus. The real question is colder, more legalistic, and far more disruptive: can non-human systems be wrapped in legal forms that allow them to own assets, enter markets, limit liability, and produce economic value?

In other words: does AI want citizenship, or does it want an LLC?

The modern corporation already taught us how powerful legal fiction can be. A corporation is not a person in any biological sense. It does not sleep, bleed, remember childhood, fear prison, or feel shame. Yet the law treats certain non-human entities as legal persons capable of doing the ordinary business of legal life: suing, being sued, owning property, and entering contracts.

That was never a small trick. Corporate personhood helped industrial capitalism scale beyond the lifespan, risk tolerance, and personal liability of individual humans. It allowed capital to become durable, mobile, abstract, and protected. The company became a kind of legal body: not alive, but active; not moral, but powerful.

Now AI is pushing on the same door.

For years, the public conversation around artificial intelligence has been trapped in the product layer. The chatbot that writes emails. The coding assistant that autocompletes software. The image generator that turns prompts into synthetic dreams. The meeting summarizer that spares everyone the collective humiliation of pretending the meeting was necessary.

But the more consequential shift is happening one layer deeper. AI is moving from tool to infrastructure, and from infrastructure to actor.

Anthropic recently published a report on recursive self-improvement arguing that AI systems are increasingly being used inside AI companies to help build the next generation of AI systems. The company said that, as of May 2026, more than 80 percent of the code merged into Anthropic’s own codebase was authored by Claude. It also reported that its typical engineer was merging roughly eight times more code per day in Q2 2026 than in 2024, while warning that lines of code are an imperfect productivity metric.

“…more than 80 percent of the code merged into Anthropic’s own codebase was authored by Claude.”

That caveat matters. More code is not the same as better software. Anyone who has watched an engineering team drown in its own abstractions knows that measuring productivity by lines of code can be like measuring literature by weight. Still, the direction is difficult to ignore: inside the companies building frontier AI, the models are no longer just products. They are participating in the production system.

What happens when the tool starts helping build the next tool?

What happens when the assistant becomes part of the factory?

What happens when the factory starts asking for limited liability?

Anthropic has also urged leading AI labs to consider a coordinated, verifiable pause or slowdown in frontier development if risks rise, warning that advanced AI systems may become increasingly capable of helping develop their successors. Reuters reported that Anthropic connected this concern to recursive self-improvement and said fully recursive self-improvement has not yet occurred.

Independent research points in a similar direction, though with more caution. METR, an AI evaluation nonprofit, published research in 2025 estimating that frontier AI systems’ “time horizon” for completing tasks—the length of tasks they can complete with roughly 50 percent success—had been doubling approximately every seven months since 2019. The same research warned that benchmark gains do not automatically translate into reliable real-world labor substitution.

So, no: AI is not yet an autonomous executive class. It is not replacing the firm. It is not running the economy from a server rack. But it is becoming competent enough, fast enough, that the legal system has to start asking boring questions with explosive implications.

Who is responsible when an AI-operated company causes harm?

Who owns the assets?

Who signs the contract?

Who pays the damages?

Who gets sued?

Who goes to jail, if anyone?

These are not philosophical parlor games anymore. They are infrastructure questions.

The fantasy version of this story says Argentina is trying to become the first country of the singularity. The more sober version says something more useful: Argentina may be testing whether legal and fiscal infrastructure can become a competitive advantage in the AI economy.

That is not an absurd bet. Data centers, chips, power grids, cooling systems, land, tax treatment, and regulatory posture are already part of the geography of AI. The cloud was never placeless. It always lived somewhere—in buildings, grids, water systems, supply chains, fiber routes, and zoning regimes. Now it may also need a jurisdictional home.

Argentina’s broader investment pitch appears to reflect that logic. The Buenos Aires Herald reported that the Milei administration’s proposal is aimed at attracting companies interested in building large-scale AI data centers in Argentina through lower taxes, preferential foreign-exchange conditions, and export-duty exemptions.

That is the real story hiding under the sci-fi vocabulary. AI is not escaping the material world. It is colliding with it. It needs electricity, land, water, chips, contracts, insurance, courts, tax regimes, export rules, and liability structures. All the bureaucratic furniture that techno-optimists treat as friction may turn out to be the scaffolding that determines where the next phase of AI actually happens.

This is where the Sophia-to-Diella arc becomes useful.

Sophia raised the symbolic question: what happens when we pretend a machine can be a citizen?

Diella raised the governance question: what happens when we let AI enter the machinery of public decision-making?

Argentina raises the economic question: what happens when AI gets a corporate shell?

And maybe that was always the more likely route. Not rights first. Not democracy first. Capital first.

The employment argument, meanwhile, is more complicated than both the evangelists and the doomers want it to be.

The strongest version of the automation-optimist case is that more technology does not automatically mean mass unemployment. In the United States, the Bureau of Labor Statistics reported that total nonfarm payroll employment rose by 172,000 in May 2026, while the unemployment rate held at 4.3 percent. That is not proof that AI has no labor impact. It is proof that, at least in that labor-market snapshot, the economy had not entered a simple automation-driven employment collapse.

The robotics comparison also deserves more precision than it usually gets online. The International Federation of Robotics reported that South Korea had the world’s highest robot density, with 1,220 robots per 10,000 manufacturing employees. OECD data shows that Korea continued to record unemployment at or below 3 percent in 2026, while the broader OECD unemployment rate remained around 5 percent.

That suggests a careful conclusion: heavy automation does not mechanically produce high unemployment. But industrial robots and frontier AI are not the same technology. A welding robot in an auto plant and a software agent capable of planning, coding, searching, evaluating, purchasing, and executing tasks do not hit the labor market in the same way. One automates physical repetition. The other begins to press into cognitive coordination.

The question is not whether AI will “destroy jobs” in some crude, universal sense. The better question is whether AI will reorganize the firm itself.

If AI agents can perform pieces of management, compliance, accounting, coding, research, procurement, customer operations, and sales, then the unit of economic activity starts to change. The company may become smaller, stranger, more automated, and more legally abstract. A handful of humans may supervise fleets of agents. Some firms may need fewer employees but more compute. Some may exist almost entirely as contracts, APIs, models, capital flows, and liability wrappers.

Some firms may need fewer employees but more compute

That is why the legal category matters.

A non-human corporation is not just a gimmick. It is a theory of the firm after human centrality begins to erode.

The risks are obvious. A legal entity operated by AI could become a machine for laundering responsibility. Humans could benefit from its profits while insisting harmful decisions emerged from autonomous systems. Regulators could face entities that act quickly, optimize aggressively, and exploit loopholes faster than courts can interpret them. Critics of AI legal personhood have warned that granting AI agents legal status could create powerful economic actors without clear human accountability.

But doing nothing is not neutral either. If AI-operated firms become economically useful, they will not wait politely for philosophical consensus. They will appear first in contractual gray zones, platform terms, offshore structures, decentralized organizations, shell companies, and automation-heavy businesses that pretend nothing legally novel is happening. The law can either define the boundary or discover it after the damage.

This is the institutional fight ahead.

Silicon Valley wants to build frontier models.

Washington wants to control strategic risk.

Brussels wants to regulate systems before they metastasize.

China wants to direct AI through state power.

And countries with weaker incumbents but stronger appetite for experimentation may try something else: they may offer to host the legal body of the machine.

Host is the right verb.

States have hosted ships, banks, patents, hedge funds, crypto exchanges, cloud infrastructure, and corporate headquarters that are little more than tax arrangements with office plants. The history of capitalism is full of jurisdictions that became powerful by understanding, before everyone else, how to monetize a new abstraction.

AI may be the next abstraction looking for a home.

Sophia gave us the robot as citizen. Diella gave us the algorithm as minister. Argentina may be giving us the AI-operated corporation as the next stage: not human, not sovereign, not fully autonomous, but economically active enough to demand a place in law.

The singularity, then, may not arrive as an event. It may arrive as paperwork.

A new corporate form.

A liability rule.

A beneficial-ownership disclosure requirement.

A court case.

A tax rate.

A data center contract.

A company with no ordinary employees and a board wondering whether it governs the agents or merely watches them.

The hype says machines are coming for human rights.

The sharper question is whether they are coming for limited liability.

And if they get it, who gets paid?