There are many possible futures for AI. Even if we cannot predict exactly which path it will take, as long as we remain clear-eyed about the possible consequences of the actions we take today, we will be able to mitigate the outcomes that could result over the next few years.
This is a link-enhanced version of an article that first appeared in the Mint. You can read the original here. For the full archive of all the Ex Machina articles, please visit the website.
The future arrived quietly, wrapped in convenience.
When they first appeared on the scene in 2025, artificial intelligence (AI) agents were simple. We used them to schedule meetings, write emails and occasionally help negotiate with customer care. We were delighted to outsource these relatively low-stake tasks to a digital assistant that was able to perform them so well. Over time, we began to rely on them implicitly.
Today, in the year 2035, Agentic AI is our primary interface with the world. These digital assistants have, over the past decade, transformed from marginally helpful tools into our virtual proxies. Not only do we use them to execute complex tasks under our instructions, we’ve grown so confident in their ability to know exactly what we want that we allow them to anticipate our needs and make decisions on our behalf. As a result, we have, in many aspects of our daily lives, completely withdrawn from daily interactions, leaving our AI agents to participate on our behalf. In the process, the world changed without our knowledge and in ways we did not anticipate.
By 2026, AI agents were negotiating our contracts, investing our money and even managing our commercial relationships, so much so that we have, by now, grown accustomed to their acting on our behalf. Today, we all have so many AI agents out in the world that it’s hard to keep them all aligned. Every now and then, one of them commits us to behave in ways that contradict what another agent promised to do, and we find ourselves in court over this.
We tried to argue that we could not be bound by our AI Agents, but the courts, in an attempt to bring some order to the proliferation of AI Agents, ruled otherwise. Now, we have no choice but to honour the commitments they make on our behalf.
The better they got at helping us in our daily lives, the more we used AI Agents to do work we once relied on human assistants to perform. While this resulted in some job market displacement, it was not because they made human skills redundant but because it was simply more efficient to have AI Agents interact with each other. Today, our productivity is no longer linked to human labour but the sophistication of the AI we use. This has created a new class system, one in which those who cannot afford advanced AI agents find themselves unable to access many of the services they need. In some countries, this is such a concern that most governments have instituted affirmative action to ensure that every citizen has at least some level of baseline AI representation. But the gap continues to widen.
In 2029, the first AI agent outlived its principal. By the time its human had passed on, it not only knew everything about him and almost every past decision he had made, it was able to converse in his voice and create videos in his likeness. As a matter of fact, for the last two years before his passing, he had used these features of his AI agent to represent him in the physical world when his illness prevented him from participating in person. After he died, some of those with whom he had been transacting did not realize for as long as six months after the fact that they had for some time been interacting with an AI agent all that time.
This forced us to think through issues we never had to deal with before. What happens to the assets of a human if they have been purchased by his AI agent after his death? What about the business decisions that it took—should they be honoured or revoked? And who owns the intellectual property that his AI agent created? Given that so many of these interactions involved a complex web of inter-operating agents, we had no idea how to shut it down or whether the AI Agent had a legal right to stay in existence. Some countries have since enacted new legal frameworks to address issues of posthumous AI agency and, in the process, redefined existing notions of inheritance, legacy and mortality.
Most couples today have created AI agents that they share—joint AIs that manage their finances, schedule their plans and speak for them as a couple. This is all well and good until the marriage falls apart, after which it is significantly more challenging to disentangle two now-separate lives from an AI agent designed to only think of them as together.
Divorce is messy enough when it involves houses, children and assets. It has been very difficult trying to figure out how to appropriately apportion the attention of an intelligent agent who knows not only the entirety of the couple’s private conversations but also their unspoken dynamics, secret fears and desires. Do we split the AI agent, forcing it to ‘forget’ half its knowledge, or allow it to remain whole but in stasis, no longer belonging to either one or the other? We found that our laws, as good as they are in apportioning matrimonial property and custody, were incapable of disentangling what is neither.
In 2025, no one thought much of AI agents. They were simple digital assistants, limited in what they could do and how well they did it. In less than a decade, this new technology has embedded itself so deeply in our lives that we are completely dependent on it. There are those who say we should destroy this technology before it completely strips us of all agency. Others argue that it is already too late for that.
This is something we should have thought about when AI agents first arrived on the scene 10 years ago—when it was still possible to shape how they behave and conduct themselves on our behalf.
