Engineering deep-dives, design decisions, and notes from building atrium.
An interactive, sourced exploration of when today's frontier AI could run locally on a phone — model efficiency, memory, hardware, and recursive improvement, with assumptions you can change.
A full audit of all 2,053 files in X’s open-sourced recommendation stack — the 23 published ranking coefficients, what the code proves, and what it cannot.
The most upvoted complaint about AI coding isn't that it's wrong — it's that the work stopped feeling like anything. The research says the flow tax is real. But it's not the model's fault. It's the pile of terminals you bolted it onto.
What Tauri + CEF actually costs: a 170MB Chromium runtime, native view plumbing, and the control you need for OAuth, CDP, and real browser panes.