


Mostly harmless • Perfecting AI for transactional lawyers • augustus.esq • wargame.esq


I could be wrong, but I suspect machines can't escape their fundamental limitation: at their core, they do exactly what they're told to do. Perhaps they can't die, but they can definitely get caught in an endless loop or get stuck waiting for a signal that never comes.

If all work is AI-augmented, the leading edge in every domain will always provide new, confounding edge cases that challenge the "old model". Even with accelerating self-improvement loops, the messy frontier will present opportunities for existential stalemates or meltdowns.

While I think general intelligence will expand beyond human scale—and most likely already has—I also think that practical, self-sustaining "omniscience" is unlikely to emerge in a neat package.

There is a debate about whether or not a "single model" will absorb all domain knowledge or whether we will always need to curate general intelligence to keep it aligned with our reality, and to fit it to specific domains and use cases.

Apophenia really flaring up in late 2026.

We read the ZDR docs of 12 AI providers. "Zero" means something different for everyone. For Tinfoil, zero is something you can verify. Can you trust ZDR policies when they are riddled with exceptions and caveats? tinfoil.sh/blog/2026-11-16-z…

Japanese sauna by Daichi Inc makes “people feel as though they are dissolving into the forest”: dezeen.com/2026/11/14/daichi…


I think the future of software is disposable, one-off, just-in-time solutions. Designing and maintaining software for a broad collection of use cases won't last much longer. Data sources and identity and authorization will remain stable, but the rest will take shape on demand.

The "ChatGPT" versus "Codex" switch serves no purpose at this point except to sow confusion.

I guarantee you that in five years, you will still need to turn up in person and fill out forms by hand—no mistakes!—in order to get anything important done in Japan.

Pics taken right before I got lost in the woods on the backside of the mountain, got spooked because I felt eyes on me, and had to backtrack over the summit, barely making it down in time for sunset.

"Chinese spy" is the new "Russian bot".

When Alice Huang is referred to as an “alleged” spy, who is making the allegation?
