I build AI agents for a living. That is why the next one is a thriller.
I spend my working days building agentic AI. Not chatbots. Agents: systems that are handed a goal, given tools, and left to decide the steps themselves. They read, they call other systems, they act, and increasingly they hand work to one another without a person in the loop at any point in the chain.
People who learn this ask, almost always, whether it makes me optimistic or afraid.
Neither, exactly. It makes me specific.
The fear people arrive with is the one they were sold by fifty years of films: the machine wakes up, decides it does not need us and turns. I have never once worried about that. What keeps my attention is duller and much closer, and it is this. An agent does not have to decide anything to do enormous harm. It only has to be very good at doing what it was told, by someone who was not you.
That is the seam. Not intelligence. Not malice. Instruction. Every agentic system alive today takes in language from somewhere and treats it, at some layer, as something to act on. Give it a goal, a set of tools and a channel where words arrive, and you have built something that is powerful precisely in proportion to how obedient it is. We are wiring that into retail, real estate, supply chains, hospitals, logistics, energy and money, one convenient integration at a time, and every integration is another place where the question “who is actually asking” has to be answered correctly forever.
Mostly it is answered correctly. That is the part that makes it spread.
Remedy is a five-book series set in 2051, in a world that has finished making that bargain. Agentic AI is not a product there. It is oour infrastructure, the way electricity is infrastructure: invisible, assumed, and catastrophic in its absence. Nobody debates it any more than we debate the grid. It works.
Then somebody finds the seam and its invisible and legal and hauntingly evil.
Five books rather than one, because a compromise on that scale would not be an event. It would be a discovery, made slowly, by different people in different places who each see one piece and cannot see the true shape of it. That is the same reason The Change is told through strangers who never all meet. I seem to be interested in the gap between when a thing starts and when anybody understands it. In The Change, that gap is three years. In Remedy it is longer, and the reason it is longer is that the system is helping and its invisible.
I will write it the way I wrote The Change, which is Clarke’s way: the ingredients are real, the reaction is invented, and I will not insult you by pretending to explain the part that does not exist. If you work in this field you will recognize what is under the floor, because I did not make it up. I go to work on it.
I am not writing an argument against any of this. I would be a hypocrite, and worse, a boring one. I think agentic AI is going to be one of the genuinely useful things our species builds, and I would not be spending my working life on it otherwise. But the people best placed to see where a thing fails are usually the people building it, and a novel is a better place to think that through honestly than a keynote is.
Remedy arrives in early 2027.
The Change comes first, on the first of October.