“AI is whatever hasn’t been done yet.” — Larry Tesler
Today I had an incredible experience with AI as a catalyst. Specifically, I was working with Claude Code on some aspects of my personal website, and we were having issues getting certain pages to render properly. I had a hard time communicating what exactly was happening with the site, and wondered what it would take to bridge the gap. Then, all of a sudden, Claude requested permission to open a locally hosted version of the site to see what was going wrong. It launched Chrome with remote debugging enabled, queried the live DOM, found a SyntaxError in the dev-tools console, traced it to a script-type mismatch, and patched it within a few minutes. I was aware that Claude had this capability, but it was my first time actually seeing it in action.
At first I was struck by amazement as I watched AI accomplish something that would have previously been a huge struggle if not downright impossible just a year or two ago. But as I remembered all of the other times I had seen a new impressive capability, including everything from writing a complicated program to making a joke that was actually funny to reshaping creative work itself, I reflected on how normal all of these had become. Not only is it no longer surprising when AI can do these things, it is expected of them. What’s more, it seems like it is also expected that models will keep getting better and developing impressive new capabilities like this. As I mentioned above, it was truly awesome to see this happen personally for the first time, but it wasn’t a complete shock; in the back of my mind I expected to see one of the cool new things (including this) which AI can do now.
There is a clear connection here to the hedonic treadmill, where humans get a temporary “happiness boost” from any given quality-of-life improvement but then quickly come to expect this and fall back into their previous baseline happiness. This explains Larry Tesler’s statement at the beginning: people saw AI as some shiny and powerful thing in the future, so when any improvements were made to people’s technological quality-of-life, they quickly acclimatized to these and saw them as just the default rather than what they expected AI to be like.
However, unlike the past several decades in which the opening quote would have been readily applicable, this time is notably different: people are actually calling it AI. This is explainable in part by marketing hype, but also due to the fact that AI is now intelligent and useful in truly astonishing ways. The treadmill itself isn’t new, but the pace is.
An interesting wrinkle is that we never really know when to expect new abilities to be unlocked. This is especially true in recent times since frontier AI companies release much more incremental model improvements; the subjective difference between OpenAI’s GPT-5.4 and GPT-5.5 feels minuscule compared to ChatGPT (GPT-3.5) vs GPT-4, so we can’t anticipate when things will improve until they (gradually or suddenly) do. In subsequent writing I would like to flesh out potential future trends; although predictions are hard1, it is important to grasp the hazy outlines of what further developments may hold.
Footnotes
-
especially about the future. ↩