The technology on this site is real. The products are not… yet.

None of the customers here are unusual. A man who lost an argument over a parking space. A woman whose ex moved on. A couple who waited too long and can't stop thinking about it. People have always wanted revenge, wanted money they haven't earned, wanted the dead back and the missing filled in. What is new is not the wanting. It is that we are about to hand every one of those ordinary impulses an intelligence that can plan, and a reach that can act — patiently, anonymously, at scale, from anywhere — for as long as the machine is left running. For most of history, the gap between wanting something and being able to do it has been friction: effort, skill, time, money, embarrassment, other people, eventually getting tired or moving on. That gap is closing.

Consider how loud the world already is: the robocalls, the scam texts, the fake reviews, the inbox nobody can clear. All of that is produced by people with limited time and limited skill. Remove both limits and the noise stops being an annoyance and becomes an atmosphere. Every scam, grievance, sales pitch, obsession, political cause, lonely person, angry customer, abusive ex, con man, corporation, hustler and crank can have something intelligent working on its behalf, all at the same time. The amount of effort competing for our attention and money can increase almost without limit. Our attention cannot.

We spend a lot of energy on the grand version of this problem (the data centers, the power and water they draw, the jobs, the question of whether the thing ever turns on us), and we should. But the grand version unfolds slowly. The ordinary version arrives the afternoon this reaches the people you already deal with — the scammer who already has your number, the neighbor who already holds a grudge, the worst person you know. None of them are going to become better people because the tool got smarter. They are going to do exactly what they do now, except tirelessly, and to more of us, and from everywhere at once.

And when one of these systems does something harmful, it may not even be obvious who did it. Was it the person who gave it the task? The agent that decided how to carry it out? The model that misunderstood what the person meant? The company running the application? The company that made the model? Someone who changed the model after downloading it? Someone who fed it an instruction from the outside? Intent, action, control and responsibility can all end up in different places.

Today, much of the practical restraint standing between us and all of it is a handful of AI companies choosing what their models will refuse. Laws are beginning to draw lines too, but they are jurisdictional and this technology is not. A rule in California does not determine what a copy of a model running somewhere else will do. A company can decide what happens on its own servers. That control ends when the weights are released and the model is run on a machine someone else owns.

We have decided, so far, to approach a global problem company by company and jurisdiction by jurisdiction, while the thing being governed can be copied and moved anywhere. We have not agreed on who gets to decide what an autonomous system may do on a person's behalf, who is responsible when it does something the person did not explicitly ask for, or what boundaries survive once the model leaves the control of the company that made it.

These decisions are being settled right now, while the tool is almost ready and becoming harder to restrict. The grand version may still be years away. This one is not. The products above are what it looks like when it reaches ordinary hands.