Vercilo is a personal publication on AI, hardware, and the systems we build between them. No team, no sponsorships, no “what we're reading” — just essays, when they're done.
Hello. I'm the editor of Vercilo — a small, slow, one-person publication about the seams of modern computing. The pieces too unfinished for a paper, too involved for a thread, and too specific for the front page of a tech blog.
I've spent the last decade somewhere between systems engineering and machine-learning research. I've shipped a database, two compilers I'm not proud of, and one model serving stack I am. I've reviewed papers, fixed memory leaks at 3 AM, and watched several “next big things” arrive on schedule and then quietly leave again.
Vercilo is where I publish what I learned in between — long arguments, mostly about why a thing works the way it does, and what would have to change for it to work differently.
Two essays a month, give or take. No newsletters-of-newsletters. No tracking. No discord. The contact details further down work; I read them.
A standing “now” page in the Derek Sivers tradition. Updated when something material changes, not on a schedule.
A long piece on solo apps, the tooling shift behind them, and what they reveal about “team-shaped” software. Currently in the second draft.
Last edit · APR 26, 2026Still trying to convince myself the gains hold up outside the benchmark suite. Notes are accumulating in a folder titled, optimistically, maybe an essay.
Updated · APR 21, 2026Lightweight tooling to grade my own drafts against a rubric I've been refining for two years. It is, very obviously, an excuse to not write the next essay.
Updated · APR 14, 2026Names and titles are deliberately omitted; the work is the only part that aged well. Roughly chronological, most recent at top.
The rules I write under. They're personal, not prescriptive — but every essay on this site has passed through them.
No ghostwriting, no editorial collective, no “contributors.” If a piece runs on Vercilo, I wrote it, and I stand behind it.
Essays land when they’re done — usually two a month, sometimes one, occasionally none. The cadence of thinking, not of publishing.
I use language models for what they’re good at — research, line edits, code review. The prose, the argument, and the mistakes are mine.
If I worked on a system I’m writing about, you’ll see it in the first paragraph. No hidden alignments.
Every essay has a public changelog. When I’m wrong, the original stays up; the correction joins it.
vim · plain markdown · build = pandoc/feed.xml · always will beD9E3 4F8A 22C1 70BDA semi-regular snapshot of what's adjacent to the work. Not endorsements — just the pile.
The Computer Boys Take Over — Nathan Ensmenger
Programmed Inequality — Mar Hicks
A Philosophy of Software Design — John Ousterhout
Updated · APR 12, 2026Speculative decoding revisited — ICLR '26
Notes on memory-bound kernel scheduling — USENIX '26
A taxonomy of long-context degradation — preprint
Updated · APR 19, 2026A short film on early IBM machine rooms — Computer History Museum
Most of Charles Mingus's late-period live recordings — while editing
A long interview with Dan Luu — in tabs, unfinished
Updated · APR 22, 2026I read every email. I reply to most of them, eventually. The best way to start a conversation is to send a long one — about something you read here, or something I should read.
A short list of the questions that arrive most often, answered once so I can stop drafting the same reply.
No, sorry. The site has no advertising and no commercial relationships, and I'd like to keep it that way for as long as it's mine to keep. If you want to support the work, the kindest thing is to read it and tell someone.
Yes — everything here is CC BY 4.0. Quote freely with attribution. For full reprints or translations, drop me a note first; I almost always say yes and I'll usually offer to review the translation.
Occasionally. I take a small number of advisory engagements per year — usually short, technical, and pre-product. I don't do retainers, fundraising decks, or general AI strategy. Email is the right way to ask.
Maybe. I do a handful of talks a year, usually in Europe and usually adjacent to an essay I'm already writing. I don't fly long-haul for keynotes. If your event is small and your audience is technical, please ask.
I keep a folder of half-formed objections — things I read or saw that didn't sit right. When the same objection shows up three times in a quarter, it usually becomes an essay. The good ones argue with me for a few weeks before I argue back in public.
It's a portmanteau I made up while looking for a domain that wasn't taken. Verso, the back side of a printed page — where footnotes and addenda live — and cilio, the small hairs on a cell that move things along. It scans nicely as a verb. That's the entire etymology; there is no clever second meaning.