Loading…
A field guide to things worth understanding. We take one idea at a time — a token, a context window, the life of a star, the hiss of an espresso wand — and try to make it legible without pretending anyone needs to be an engineer to follow along.
Looking Glass is a small, editorial site of considered explainers — pieces about the mechanisms and systems most people live alongside without ever looking inside. The first series is about modern AI: what a token is, what a context window does, how the frontier models actually differ. The archive will wander from there — into physics, mechanical engineering, adaptive software, anything whose working parts reward a closer look.
It is not a newsletter, a benchmark tracker, or a tutorial. It’s a reading room. Each piece is researched from primary sources, written in long form, and paired with a visual that does work the prose can’t. Every claim is cited. Nothing is scraped or re-heated from someone else’s copy.
Borrowed from the long magazine piece and the well-made diagram. Slow by design.
Looking Glass is made by Coalescence Labs, a small studio interested in the space where careful writing, working code, and considered design overlap. We build the kind of things we’d want to read — or use — ourselves.
Our working principles are simple. Cite the primary source, always. Write our own prose. Ship a piece when it’s ready, not when a calendar says so. Treat the reader as an adult with a little time and a lot of curiosity.
Privacy · We use Vercel Web Analytics for anonymized page views — no analytics cookies, no ad pixels, nothing that follows you across the web. We also log two coarse signals: when someone successfully sends the idea form, and when a reader opens an external citation from a piece (only the destination hostname, never query strings or page titles). If your browser sends Do Not Track or Global Privacy Control (the Sec-GPC header), we skip those tallies for your visit.
If any of that resonates, we’d love to hear from you — about a piece you’d like us to write, a detail we got wrong, or work you think we might be useful for.
The archive is grouped into series — clusters of concepts that share a subject. The first series, On language models, collects the AI pieces. Each new series will open its own room.
New pieces land when they’re ready. There’s no cadence to commit to, no newsletter sign-up nagging you in the corner. The index is the roadmap. Bookmark it; come back when you have time to read.
How modern AI systems read, remember, and differ — the first concepts in the archive.
Arrives when it’s ready.
The best Looking Glass pieces tend to start with someone else’s curiosity. If there’s a mechanism, a system, or a phenomenon you’ve always wondered about — tell us. We’re after the idea, not a pitch.
Questions, corrections, or commissions → coalescencelabs@gmail.com
Built with care in 2026. Set in Instrument Serif and Geist. Verified and maintained by the studio.