I make notebooks by hand, but I also built AI for years, and I enjoy automating everything around the creative, pleasurable part of the work.

My cutting machine isn’t the fanciest, and I struggle to align it. If it isn’t aligned properly, the pages don’t line up, the notebooks come out in different sizes, and the corners can’t be rounded cleanly. It has wasted hours of my time. So a few months ago I 3D-modeled a small piece to align the machine. It worked well, but then I needed more sizes, and going back through the 3D modeling and the printer settings to optimize everything wasn’t fun anymore.

Naturally, I thought: could one of my agents take over the whole workflow? Improve my design, convert it to the printer’s format, optimize the settings, and print it? So I connected everything, and now I have three perfectly sized alignment tools for my cutting machine, in the real world. I did nothing but describe my problem to an AI agent and approve the print.

This taught me that AI already interacts with the real world, even without robots. What that means, I’m not sure. And it isn’t even the part that worries me most.

Some of the things I work on I don’t want in the cloud, so I have a machine powerful enough to run small AI models, around 30 billion parameters. These models are great at very specific tasks, but they need a lot of alignment and guidance to do things properly without getting lost. Again, I thought: what if I used stronger models to tune and manage the models I run locally?

So I tried, and… it worked.

The stronger model built the test, had each candidate write forty of my morning digests, graded them, and wrote up which one should run. I read a few of the digests, without knowing which model had written them, and picked my favorite. It matched the AI’s choice 83% of the time; an AI made another AI stronger at what it needed to do.

I know what this is: one person, one machine, a hobby-sized experiment. It proves nothing about where this goes. But it is the first time I have seen it happen on my own desk, and once you have seen it, it is hard to believe it stops there.

What if we push this idea? What if we reach an infinite loop? What does it mean for us, for entrepreneurs, for investors? What should we work on, if the problems we solve are temporary? Should we invest in businesses we suspect won’t survive the next five years?

Of course, we can invest in what powers AI. But are the limits we see today relevant for tomorrow?

Take quantum computing. The promise is a machine so much more powerful than today’s that it could run AI at a fraction of the computing power we burn through on current chips. If that happens, “what powers AI” is not the same thing anymore. If I take seriously the idea that AI gets much better at research, why would I expect the obstacles to useful quantum machines to fall at the slow pace I’m used to? There is already a small, measurable example. On Google’s Willow processor, reinforcement learning reduced the logical error rate by a further 20% after experts had already calibrated the machine. The system learned to improve the control of a physical machine.¹ A quantum computer is not automatically a faster machine for every task, and any speedup still has to survive the cost of actually running the computation.² But that result makes it harder to treat today’s engineering difficulties as a stable measure of the future.

And what about efficiency? What if AI figures out ways to run itself more efficiently? AI has started to solve problems we hadn’t been able to solve. This month, OpenAI published hundreds of mathematical research results produced by one of its models, many of them as proofs a computer can check.³ Are we confident that we even know what the problems are, once something so much more able than us is doing the research?

Here is the one thing I do believe. Everything that can be handed to the machine will be. It is what I already do at my own desk: I hand over the aligning, the business operations, the settings, and I keep the part I enjoy.

So, here I am, writing in a notebook I made myself, with my own hands, and I’m thinking that maybe this is all we truly have left. Our ability to enjoy the present moment, to be creative, to interact with other humans. I am not sure I know what to think. But I will keep writing, because I write for myself, so I can understand what I think a little better.