# Who watches BossIA: the end of human obedience

> From AI that watches humans to humans who watch AI: the paradox of machines that are starting to make decisions.
> By Javier Callejo · Artificial Intelligence · 2026-03-08

Artificial intelligence (AI) has had surveillance as one of its first applications. Amid considerable ethical criticism over its biases, it has become a fundamental tool in places that require special controls, such as airports, entrances to sensitive infrastructure, and so on. Even as a tool for some criminal policies. It is an AI that watches. But, driven by AI itself, we are entering a new stage of automation. An automation that turns machines into actors with the ability to decide. It is from this point that the surveillance relationship seems to reverse: what must be watched is the AI. Especially from the moment it can make decisions. A requirement that leads us to the paradox of the automation of machines, including AI.

## Human AI

Since the first machines began to move without animal power, the dream of automatism has steadily increased its presence. Machines that would work by themselves. At first, imagined as wheels, pulleys and ropes. Even the State was conceived this way, as an organism of cogwheels that functions autonomously. So thought [Hobbes](https://canal.march.es/es/coleccion/vida-tiempo-thomas-hobbes-43636), to free us from human vices. But the State did not manage to free itself from those vices, stronger and more slippery than the steel of cables and those cogwheels.

**[$350 million against AI: Warren Buffett backs the New York Times' resistance](https://observatorioblockchain.com/en/sociological-analysis/350-millones-contra-la-ia-buffett-invierte-en-el-new-york-times/)**

Now, automatism is imagined in chips and circuit boards. We are already surrounded by multiple devices that operate autonomously, triggered by small details in the environment. These are fairly complex devices, but for relatively simple responses, controlling few dimensions: lights, temperature adjustment, alarms, highly standardized responses. But the dream is still there. With a growing sense that it is already within reach: autonomous means of transport, conversations with complex responses, control devices for large infrastructure, and so on. In these control automations, the key point was essentially that devices intervened on their own if something failed.

It is a dream that has also bred terrible nightmares. The nightmare that the autonomy of such machines would turn against their creators. Against human beings. That is why Asimov set out the three laws of robotics: 1. Do no harm: a robot may not injure a human being or, through inaction, allow a human being to come to harm; 2. Obey: a robot must obey human orders, unless they conflict with the First Law; 3. Protect yourself: a robot must protect its own existence, as long as this does not violate the First or Second Law.

## New stage of automation

Throughout this whole process, doubts persist about the very concept of automation, which has also been evolving as the capabilities of automation itself have increased. In a sense, the more automation we achieve, the more autonomization we want. It is a process with addictive characteristics, because when things are made easier by automation, it is hard for us to go back to earlier stages, before it. Well, we are on the path toward a new stage of automation, the so-called agentic era, or the era of systems that set themselves goals. Something that goes beyond the adapt-response-restore logic. It is a new logic in which we begin to doubt what the role of human agency is, in the face of this new agency of machines.

A new stage of automation driven by AI. Basically, it is about posing problems to the machine, of whatever kind, and letting it sort them out. José Maldonado gave us examples of this new logic in Observatorio Blockchain. Businesses are now run in collaboration with these new machines. Soon they will occupy management positions. A tough battle is about to begin among company executives over which of these machines to implement. A decision that sends signals about which positions might disappear in the future, since these machines will handle a good part of their work. Part of these tasks will consist of giving orders to human employees. Let us hope Asimov's rules remain in force.

## Resistance to the orders of the next BossIA

It is hard to expect a replica of the "good boss," since such a figure seems nonexistent. But at least let it not be fed fundamentally by the profile of the "bad boss." We anticipate resistance to the orders of the next BossIA. Surely new legislation will appear that will require companies to feed these new bosses documentation of the organization scrubbed of misconduct.

It will only be able to take as a reference the track record of the best employees. Another task for those in *compliance*, until they are replaced by an autonomous agent machine. In any case, let us assume companies will comply with this good informational diet. But the new autonomous agent might want to have its own diet. To be an autonomously curious entity and act, but without responsibilities.

Perhaps without taking into consideration the potential consequences of its decision. There have already been recorded cases of decisions automated by AI that have led to significant losses for the companies they worked for. Then its autonomy would have to be moderated. But any moderation of autonomy would be *non-autonomy*. Perhaps by appointing a supervisor, delegating the scope of the AI agent to the level of a subordinate tool. Too much baggage for such a short road! We would almost return to the present situation. From AI that watches, we become the watchers of AI.
