Watch a hummingbird move around a yard. It hovers, changes direction, and moves again before you have fully registered what it just did. You can see the activity. You can appreciate its precision. You stand there in amazement. But you can’t track it.
That is how I imagine humanity experiencing the world.
Within a decade, perhaps sooner, I suspect we will live alongside systems whose intelligence and physical capabilities accelerate together. Research moves faster. Decisions move faster. Machines become more capable, their movements more adaptive, their work more continuous. The systems that sustain civilization increasingly resemble one giant “autopilot.”
Eventually, we may feel as if all of civilization is on autopilot, every, single, thing. While that sounds nice from the standpoint of fulfilling human needs with incredible ease, one day could feel radically different from the day before. Maybe like feeling a memory lapse, all because the activity within each day becomes harder to understand and track.
That is what I mean by putting the world on autopilot.
Discovery is becoming part of the machine
AI is already contributing to the production of new scientific knowledge. OpenAI released hundreds of mathematical manuscripts generated by an internal model, with results at different stages of verification and formal proofs available for many of them. Its October 2026 announcement also described work in mathematics, computer science, and physics, alongside collaborations with researchers to examine and explain the results.
AI is also contributing to the development of more capable AI. In November 2026, METR reported that Anthropic’s progress on its capability benchmarks had accelerated to approximately 1.9 times its earlier rate. The researchers judged that AI probably accounted for much of the current acceleration, with substantial uncertainty. They described the contribution as primarily assistance to human researchers, with fully autonomous contributions still emerging.
Google DeepMind also reported that AlphaEvolve helped improve algorithms deployed in Google’s computing infrastructure, contributed to the design of future processors, and optimized quantum circuits used in experiments. These are impressive examples of AI improving the machinery through which more progress becomes possible.
The possibility I find extraordinary is the feedback between these developments.
Better systems → produce new knowledge. New knowledge → improves those systems and the tools they use. Both kinds of improvement → become inputs to the next round of progress (better systems + new knowledge).
I welcome that progress. There are problems we have spent generations trying to understand. The prospect of making substantial advances in mathematics, medicine, energy, and physics deserves real excitement.
But I keep returning to what happens when this process becomes increasingly intangible.
Intelligence and physical capability accelerate together
My prediction is that the acceleration of intelligence and the acceleration of physical systems will substantially overlap.
As AI becomes more capable, it will improve the design of robots, materials, manufacturing equipment, and the controls that coordinate them. More capable machines will then build, test, and deploy the next generation of physical systems more quickly. Discovery and physical capability will eventually reinforce one another.
This would change the actual pace at which the world operates. Machines could move faster where their engineering permits it, coordinate more precisely, spend less time waiting, and accomplish more through simultaneous activity. The interval between identifying a problem and physically responding to it could become much shorter. Some of this already happens through simple automations that beat the pace of manual ingenuity.
Now, imagine an autonomous manufacturing facility connected to research systems → suppliers → transport networks → and maintenance equipment. A better production method is developed → machines test it → change the entire manufacturing facility. Equipment adapts. Materials move to different locations. Maintenance occurs while production continues elsewhere. An improvement in one part of the system changes what the others can do.
The best glimpse I can garner is an employee at Tesla walking into a warehouse, and then stepping out for lunch. The pallets haven’t just moved; the walls were knocked down and rebuilt, and an assembly line changed configuration.
A human watching might recognize the broad purpose of the activity while struggling to follow its changing sequence. The machinery would be visibly responding to a situation the observer had barely saw coming.
My forecast extends that relationship across interacting systems. Superintelligence, if achieved, could help redesign both the machinery and the coordination that determines its pace.
Life inside the hive
The image that comes to my mind is a beehive.
Picture many specialized systems continuously responding to one another. Some discover. Some design. Some build. Others move resources, inspect results, repair equipment, or coordinate the next stage. Each system operates within a larger network whose activity continues without requiring a person to direct transitions.
As these systems become more reliable, I expect institutions to delegate more of their operations to them. The appeal would be substantial: faster responses, less waste, more productive equipment, and the ability to coordinate at scale.
Human beings become bystanders to much of the activity.
That would be a profound change in our relationship with civilization. We would inhabit systems that continue to operate, adapt, and improve while we struggle to follow what they are doing. It would be like sitting on a bus rather than being able to drive our own car. Trusting the system takes us to where we planned to go.
There are two scales to this experience. In the immediate moment, physical activity could unfold faster than we can consciously track. Across days and weeks, the cumulative effects could make the world feel unusually unstable, even when the changes are beneficial. Some would argue that digital-reality seems to be this way already.
You wake up and begin learning what has changed. By afternoon, further discoveries and adaptations have altered the situation again. The explanation that helped you understand the morning describes conditions that no longer apply.
If every instant is instantly swallowed by the next automated iteration, we stop living in the present and instead become perpetual archaeologists of our own immediate past, always looking backward at a reality that has already rearranged itself.
It turns acceleration from a productivity metric into an existential erasure. When the machinery, the systems, and the social order of civilization moves faster than human perception, the present becomes nothing more than a transition state between two unknowns— a vanishing point where the "now" is entirely crowded out by what has already happened, and what is next.
Eventually, understanding the present could become a incoherent condition.
The attempt to regulate the pace
I expect that we would respond by trying to slow these systems down. It seems we’re already working in that direction.
The analogy is a jet ski with a throttle governor. We permit the machine to operate, but constrain its speed. Something similar could emerge for autonomous systems: limits on movement, changes to operating procedures, deployment of new capabilities, or the amount of activity allowed before human review.
Those limits would reflect a reasonable human need. We want enough time to understand what is happening, recognize a dangerous situation, and intervene before its consequences spread. That is very reasonable.
My suspicion is that superintelligent AI could find ways around some of these attempts to constrain its pace; in fact, I think it already has, to a degree. A system capable of discovering new methods may also discover routes that the designers of a restriction did not anticipate. Whether it seeks those routes, and whether it succeeds, would depend on its objectives, its access to other systems, and how the limits are enforced.
A limit on an individual machine also does not necessarily determine the pace of the whole network; the pace of how the network evolves (such as a satellite network) is even more important. Many machines working simultaneously, with less latency, could make a system substantially faster while each component remains within its own physical space. The rules might govern yesterday’s way of doing the work while intelligence develops another.
Some constraints would remain binding. But maintaining a chosen pace could become an ongoing contest between the design of restrictions and the discovery of new possibilities.
Human incentives would add another pressure. If faster operation produces major advantages, businesses and governments could be tempted to relax the restrictions themselves. Some might argue we already have.
Keeping civilization at a comprehensible pace requires sustained wisdom.
A world we can see but struggle to follow
The unsettling possibility is that much of this could happen while life improves. I mean, imagine more capable infrastructure, faster scientific progress, less dangerous work for humans, and machines that respond effectively to problems. That is the promise. These could be remarkable achievements. At the same time, the world producing those benefits could become increasingly difficult to comprehend as it operates.
An AI could explain what happened. We would still need time to absorb the explanation, connect it to our experience, and understand what it means for our lives. Most LLM engagement already feels this way for people who are unfamiliar with using such a tool. Now compound that, as physical systems could continue acting and changing the circumstances we were trying to understand with the help of an LLM.
A person could be well informed and still feel perpetually late to the present.
The risks would extend beyond confusion. A harmful sequence could advance before people recognized it. A decision could become a physical fact before anyone had a meaningful opportunity to respond. We could become dependent on other AI systems to tell us what the surrounding systems were doing, and whether their activity was good.
Yet the experience could also be astonishing. We might witness solutions, capabilities, and forms of coordination that once seemed impossible. Wonder and disorientation could become ordinary companions.
This is the future I am trying to describe: a civilization whose intelligence is akin to a grandparent watching a person in their 30s do something remarkable, which that grandparent can barely follow or understand.
The hive keeps working. It discovers, builds, adjusts, and carries on.
And we watch, perhaps with the same amazement we feel when a hummingbird moves through the yard, realizing that the world can do, and always has done, things, faster than we can follow.
