Writing.

An operating system is the thing that makes a computer feel like one machine instead of a pile of components. It schedules attention, routes input, holds memory, and — most of the work, the part you never notice — keeps the seams between processes from showing. When the seams show, the machine feels broken. When they don't, the machine feels like one thing you can think with.

I think most software-with-AI today is a pile of components. A chatbot here, an agent there, a memory store that forgets, a tool that runs once and exits, a deploy pipeline that has never met the agent that uses it. The seams are everywhere. The user — usually a single operator trying to get real work done — spends most of their day routing between systems that don't know each other, paying a context-switching tax that is both unmeasured and enormous.

An operating layer is what would happen if you took the pile and made it one machine. Not in the AGI sense — that's a different conversation. In the operating-system sense: a shared substrate where memory persists across agents, where tools are first-class participants, where deploys and observability are part of the same surface as the chat that triggered them. The seams disappear. The single human at the keyboard starts to feel less like an air-traffic controller and more like…