Why 2028 still
feels possible.

I had put a flag in June 2028 for artificial superintelligence. Looking at it again, that made something I’m still thinking through look more settled than it is. I’m comfortable saying 2028 feels possible to me. I’m less interested in pretending I know the month.

The reason it still feels possible is the speed of change I’m experiencing when I use these tools. I keep finding myself reconsidering what I thought would take longer. That doesn’t give me a reliable timetable, but it does make me reluctant to dismiss a short one.

The part I’m watching most closely

By ASI, I mean clearly superhuman performance across essentially all economically relevant computer-based cognitive work, including AI research and development, with recursive improvement driving progress faster than we expect.

The recursive part is what keeps pulling me back to the timing. If, during 2027, AI becomes much better at doing the research and engineering that improves the next generation of AI, what does that do to the pace? And what happens when those improved systems can do more of that work again?

I don’t know how quickly that loop could develop or where it might stall. But if it becomes substantial in 2027, I don’t see why I should rule out something meeting my definition in 2028. That’s the possibility I want to keep watching.

Revenue could follow a different timetable

I’ve also written about the possibility of a financial correction as spending runs ahead of returns. I don’t think that automatically settles the capability question.

A system could become extraordinarily capable before businesses have worked out how to reorganise around it, before customers are paying enough to justify the investment, or before the economics of providing it widely make sense. Those things may happen at different speeds.

Of course, funding and infrastructure still matter. They affect what can be built and run. My point is simply that revenue taking longer to catch up would not, by itself, convince me that ASI in 2028 had become impossible.

An open question I’ll keep returning to

I’m not claiming this is a unique theory. Plenty of people are thinking about acceleration and recursive improvement. This is my own attempt to make sense of what I’m seeing and what it could mean for the world I’m building in.

I’ll be watching whether AI can carry research through to useful, demonstrated improvements; whether each generation meaningfully helps create the next; and whether the cost of getting that work done falls enough for it to happen more widely.

For now, 2028 feels possible to me. It’s one I’ll be watching closely, and writing about as my understanding changes.

This is a personal reflection on a possible timeline. It replaces the fixed June 2028 position I first published on 11 September 2026.