listening takes attention. if you’re focused on work and the music is just on in the background, are you really listening to it? i’d ask - what music do you have on while you work?
yarlson retweeted
Horrifying to see more than 10,000 Ukrainian civilians disappear into the Russian prison system. This is the utterly grim story of one cell: newyorker.com/magazine/2026/…
This does Coolify a disservice. A major upgrade shouldn’t be one huge PR. Use a v<N> branch, merge small reviewable PRs into it, keep it synced with master, then merge or fast-forward it back.
Anecdotal eval, the new religious doctrine of AI benchmarking.
I can't recommend GPT-5.6 for serious production code at this point.
I gave the exact same task to GPT-5.6 Sol Ultra and Opus 5 Ultracode. It was a large refactor designed to simplify the code and make the modules easier to extend in the future.
Where Sol won:
– Sol took 7h30 to finish. Opus took 9h.
Where Opus won (by far):
– Sol wrote +4,440 -500 lines of code across 41 files, including +1,721 -329 (net +1,372) lines of implementation and +51 -88 (net -37) lines of comments. The rest were tests.
– Opus wrote +977 -254 lines of code across 23 files, including +187 -203 (net -16) lines of implementation and +157 -41 (net +116) lines of comments, mostly useful "WHY" comments.
– More than half of the tests Sol wrote covered internal helpers it had created, mostly implementation details. Opus stuck to user-facing API tests.
– Sol introduced 2 regressions. Opus introduced none.
– Sol changes caused significant performance regressions, from 12% to 70%. Opus maintained the same performance.
– Sol tried to solve everything at once. Opus created 2 follow-up issues, which, after reading them, I believe was the better decision.
I’ve had this idea for several years, but, first I built a service template, then my own template/UI system, then an AI harness to help me use my own stuff, then a framework for building harnesses, and somewhere in between, I started to build my own programming language.
So I set out with the goal this year to create a linux distro from scratch for a user of one, me. I thought AI could help me. Current status: I built a tool, to build a tool, to help me use AI to build a distro. So what have I accomplished. Nothing yet. The typical, before I do something I first have to write a new operating system.

















