An Unfinished History of Intelligence, Part 7: The Next Chapter
Every earlier part closed by naming what the new mind could not yet do. This one cannot, because the story is not finished. We are inside the chapter now, not looking back at it.
Not long before this series reached its end, a research group that has spent years measuring what AI systems can do published a number that is easy to state and hard to absorb.
They gave frontier models the kind of multi-step software task a skilled engineer might work through over the course of a day, and measured the longest such task a model could complete on its own, unaided, with the flip of a coin’s reliability. Then they compared it against the same measurement taken every few months, going back years. The pattern was a clean exponential. The length of task an AI agent could carry out autonomously, at that even-odds reliability, had been roughly doubling every seven months for half a decade, and in the most recent stretch the doubling had sped up to something closer to four. A capability measured in seconds in 2019 was measured, by early 2026, in many hours. A machine that not long ago lost the thread after a few minutes of independent work could now hold a goal, across many steps and many decisions, for something approaching a full day of the work of a human professional.
For six chapters, this series has been a history: a story about the past, told from the safety of knowing how each episode turned out. Every part ended the same way, by naming what the new kind of mind still could not do, because from where we sit we can see exactly what came next and exactly which limit the following chapter would lift. That vantage point is the luxury of writing about things that have already happened.
This chapter does not have it. The doubling curve does not describe the past. It describes a line moving under our feet, right now, and where it goes is not a fact to be reported but a question that is currently open. So this last part cannot be a history in the way the others were. It has to be something stranger: an account of a story caught in the act of being written, by processes we set in motion and do not fully control, toward an ending nobody has seen. This is why the series is called what it is. We have followed the thread from the first cell to a machine that models minds, and we have arrived, not at a conclusion, but at the present, which is the one place the thread has not yet been laid.
What we can do, honestly, is name the four frontiers where the thread is being extended as you read this, and be clear about what is settled and what is not.
