In the early days of computing, “using a computer” wasn’t really about using it at all. Honestly, it was about keeping it working long enough to do something useful (anyone who has seen the film “Hidden Figures” will know what I mean). In the very early days, you actually had to understand the hardware. Not in a conceptual sense—in an electrical engineering sense. You needed to know what might fail, how to diagnose it, and how to coax it back into operation. A multimeter and an oscilloscope were not optional tools. They were part of the interface. If something didn’t work, the problem wasn’t abstract. It was physical. You went looking for it.
Even once the machine was running, the next layer wasn’t much friendlier. You didn’t ask a computer to do something. You described it in a language that the machine could interpret, with very little tolerance for ambiguity. The burden of translation sat entirely with the human. If the result wasn’t what you intended, it was because you had not been precise enough in describing it.
That didn’t change because computers suddenly became more powerful. It eventually changed because the interface improved. Over time, layer after layer was added between the human and the machine. First, operating systems reduced the need to manage hardware directly. And programming languages became “higher level” and abstracted away the details of execution. Then, graphical user interfaces removed the need to remember keyboard commands and the interface became visual and auditory. Each step made the system more accessible, more like interacting in a way that humans were naturally comfortable with. Each step allowed more people to extract more value from the same underlying capability.
But even in the age of handheld, touch screen, voice-command-able computers every stage, there was still a requirement: you had to learn how to speak to the system in its own terms. You really had to have at least some sense of what the computer could do–and how it had been designed to do it. After all, how many of us use even a fraction of the customization options on our phones. We know it’s there, but honestly it is just too much work to figure out how to use it.
So, there is still a requirement to know something about how the computer in your hand operates in order to really get the most out of it. That requirement has always acted as a filter.
In the case of really powerful computers, it hasn’t just determined how efficiently people could use computers. It has determined who could use them at all.
Which has meant that for most of the history of computing, a large portion of its capability was effectively out of reach for most people. Not because it didn’t exist. Because it was impractical to access.
If you weren’t comfortable working at a command line, writing scripts, managing environments—then a significant amount of what a modern machine could do simply wasn’t available to you. You might benefit from it indirectly, through applications built by others. But you couldn’t reach into the system yourself and make it do something new.
And this meant that it looked like the demand for all the compute power that was available was limited. Not because people lacked ideas. Because they didn’t know there was a way to act on those ideas.
There is a useful analogy here, and it has nothing to do with software.
A hundred years ago, a miner needed to be skilled with a shovel. The work was physical, repetitive, and dangerous. Output was constrained not just by the size of the deposit, but by the amount of effort a human being could reasonably apply.
Then machinery arrived. The miner no longer applied force directly. He controlled it. The skill shifted—from physical endurance to operating equipment. Output increased. Risk decreased. Later, automation arrived. Now the miner supervised systems that did the work. The role shifted again—from operating machines to coordinating them.
And eventually, the highest leverage came from, and is coming from, programming those systems so they require less supervision altogether. Now, a single individual can influence the output of an entire operation—not by working harder, but by defining what work should be done.
At each step, the underlying resource (what needed to be taken out of the ground) didn’t change. What changed was the interface between the human and the work. And with each change, something subtle happened. The effort didn’t disappear. It moved. Away from direct execution and toward defining intent. Humans moved away from the demanding, dirty, dangerous, and repetitive tasks to ones where they were solving problems and implementing solutions. We have always done that when we could.
Well, the same pattern has been playing out in computing.
For decades, enormous processing capability has been available. Graphics Processing Unit’s (GPUs), in particular, have offered levels of parallel computation that far exceed what most people ever use. But for the majority of users, that power was effectively locked behind layers of complexity. The common ways to “use” a GPU were through pre-packaged applications—games, rendering engines, or, for a while, cryptocurrency mining.
The capability was there. The access was not.
I would argue that the most profound effect of large language models (what we usually refer to as AI) have changed.
Not because these “AI engines” are “smarter” in some sense. And not because they can write code or do other things in a way that replaces a human. But because they can interpret intent expressed in plain language and translate it into actions that the system can execute.
For the first time, a user can approach a computer and describe what they want in their own terms, and with almost no prior knowledge of hardware, operating systems, coding languages or software—and suddenly find that with a bit of planning, and probably a bit of trial and error, they can actually create something that approximates that intent.
Not perfectly. Not reliably in all cases. But well enough to be useful.
And that “well enough” matters.
Because it means it is no longer necessary for someone to fully understand the system before they can begin to use it. It allows action to precede expertise. Understanding can follow.
Oh Boy! Is that ever a different operating model. And it has a second-order effect that is easy to miss. As soon as people can reach the power that is already there, more and more of them start to use it.
Not a little. A lot.
For years, computing demand has been constrained by the difficulty of accessing it. People learned—often unconsciously—to stay within the boundaries of what they could realistically execute. They didn’t attempt certain things, not because those things lacked value, but because the effort required to achieve them was too high.
Often the barrier was that they would have to explain what they wanted in a great detail to another human, and probably iterate that design according to a schedule that wasn’t fully under their control. You didn’t just try things to see if they would work. You only went looking for a competent coder when you knew you could make their time worthwhile
Remove that barrier, and the constraint shifts. Now it is about finding out if something is possible. It is no longer about whether it is worth asking for. And, it turns out that the number of ideas that people can dream up far exceeds the number of things it was worth asking for, when you had to pay someone else to try them.
This is why I think the current discussion about an “explosion in computing power” is slightly misframed.
The hardware is improving, certainly. But I don’t think that is the primary driver of what we are seeing. What we actually are seeing is mostly explosion in demand.
And that demand is being driven by access. Not by capability.
You know, we have seen this pattern before. It used to be hard to fly across the Atlantic. Not impossible. Just difficult enough that it required planning, resources, and a certain tolerance for inconvenience and even discomfort. So, most people didn’t do it. Not because they didn’t want to go, but because it wasn’t worth the effort.
But now, no one thinks about it twice.
And as a result, we don’t just have easier flights. We have more of them. Entire industries have formed around the assumption that crossing an ocean is routine. Business models, supply chains, and personal expectations all reflect the fact that movement is no longer the constraint it once was.
In other words, the demand didn’t appear. The barrier disappeared.
I think the same thing is happening with computing. For decades, a great deal of capability sat behind layers of translation—technical skill, specialized knowledge, time, and effort. Most people stayed within the boundaries of what they could realistically execute. But now those boundaries are shifting. Not because the machines suddenly became more powerful. But because more people can actually reach the power that was already there because we now have an interface that we can, literally, speak to in our own language–natively, with all of our verbal tics, expressions, and aphorisms. So suddenly many more people are finding ways to access the power in the silicon chips that have been sitting on their desks or in their pockets for a long time.
This has implications are not about whether AI is replacing human intelligence. It’s not. But it is replacing human effort, in the same way that a backhoe replaced a gang of folks with shovels. The presence of a truly natural interface with a computer means that the human input is less about knowledge and skill and more about ACTUAL intelligence. It allows humans to move from execution to judgment. From process to problem definition. From knowing how to do something… to deciding what is worth doing.
The early computer user needed to understand the machine. With the advent of AI language models, the modern user needs to understand the problem. But it also means that the limiting factor is no longer the system. It is the person asking the question.
We did not suddenly create a new class of powerful machines to take over our lives and replace us in the labour force. We made it possible for more people to reach the power that was already there. And as soon as that became possible, we discovered something that had been hidden for a long time.
We want a lot more of it than we thought.