May 22, 2026
Universal Machine
Last week I asked Claude to do something it had no business doing. "Can you edit video?"
I had ten short video clips that needed to be cut into one piece with subtitles, in a specific style. Normally I'd use Premiere for a task like this, and spend an hour or two perfecting every frame. I asked Claude to do it instead. The first attempt crashed. The second one didn't. It read its own failure as context, picked a different approach, parsed the script into subtitle layers, stitched the video files together, and exported a clean MP4. A few minutes total.
What it did wasn't magic. It wrote the tool it needed to do the task and then ran that tool. There was no traditional software in this story. There was a chat and a task, and somewhere in the middle the chat became the software for the duration of the task, then stopped being software again. I don't have the right words for this yet.
I wished, for a long time, that technology would stop evolving. The wish came up every few years, around the time some tool I'd just gotten good at became the wrong tool. Counting back: cameras moving from film to digital tape to memory cards. The cable modem that made my HTML skate video site shared with friends possible in a pre-YouTube world. The smartphones that reinvented content distribution and consumption. Specialized software like Sketch coming for Adobe products. Democratized tools like Figma pushing fresh ways to work in design. Each shift looked smaller from inside than it did from outside, but that's five rebuilds in twenty years. The general adoption and integration of AI is the sixth.
The pattern in each of the previous five was the same. A tool stack that felt permanent stopped feeling permanent. The people who'd built their identity around the old stack mostly stayed there. The people who'd built their identity around the work moved. The instinct to dig in is the tell. So is the instinct to chase every new tool. What worked the other five times was something else: stay with the work, not the tools.
This shift is different. It looks like an infinite instance of Turing's universal machine that we're now personally responsible for. It feels like overwhelm, but simultaneously great potential. The new tool is the absence of a tool at all times, until you need it. Then it will be only what you need for the duration you need it. "It is possible to invent a single machine which can be used to compute any computable sequence," Turing wrote in 1936. It took ninety years for the consumer version to show up, but it's here.