Software is changing because of AI. But we need humans in the loop more than ever.

There's a feeling of change in the air. Even people who haven't used AI much are beginning to sense that things are different year to year, month to month, and sometimes even day to day. The monumental geopolitical events of today and the recent past have diverted attention from the rapid advances in AI and automation, but they haven't slowed it down. For the last two years there's been a rough consensus that the power of AI has been doubling every six or seven months, measured by AI agents' ability to carry out tasks autonomously for a measured period of time. Recent measurements suggest it's accelerating. To get to that level of improvement, other aspects of AI have been growing even faster; much faster.

The moment a technology starts doubling consistently, it is on an exponential path. The striking thing about exponential growth is that we don't see it, and certainly don't feel it, until much later in its growth cycle. This is what happened with Covid. Retrospective analysis showed that the virus's growth was exponential months before a significant number of observers realised it. This is a feature of exponentiality, not a bug, because exponential change is deeply unintuitive: it doesn't feel right. So we either don't notice it, or dismiss it if we do.

And when we do notice it, it's often too late.

But, too late for what?

In practice, it means that some or all of the fast-changing technologies have improved more than we expected. Sometimes it takes even experts by surprise. But once you get into the habit of focusing on accelerating technologies, and largely ignoring the traditional linear ones, you start to see things more clearly.

It's like joining a motorway. It's terrifying for new drivers because you're expected to merge with traffic that's moving much faster than you. The instinct is to slow down, to be cautious, but that's how accidents happen. The trick is to accelerate quickly so that your speed matches theirs. When you do that, it all makes sense: your relative speed to the other cars is close enough that you can judge the gaps and find a space you can easily move into. Suddenly it all makes sense. Suddenly, you're part of the system. That's what you need to do with exponential technology, which, right now, is everywhere. This ability to bridge the linear and the exponential defines the role of experts in the exponential age.

Not everything is on this path. Software and computers ("compute") have been on an exponential path for many years. But with AI, this is different. The change is faster and deeper, and it breaks through barriers that were once (maybe three years ago!) thought to be impenetrable. A chair is still a chair. A jet engine is still a jet engine. And a complex, multinational organisation is still a complex multinational organisation.

And perhaps surprisingly to some, those who design, architect, organise, test, verify and sign off projects will still be people. They will still be experts, but with a newly formed appreciation of acceleration and scale. There's hope and even cause for optimism for humanity. We owe it to ourselves to understand the new world, which is a different world every day.