What an AI-native operating system actually is
A copilot bolted onto a suite is not an AI-native OS. The difference is which layer owns the data model, schedules the work and enforces the rules.
- AI-native OS
- system of action
- architecture
Blog
Design notes from building an AI-native OS: what the term actually means, how agents are governed by policy, and why one data model beats eleven integrations.
Policy, not prompts
An agent allowed to act needs boundaries a system can enforce, not instructions it can be talked out of. What a policy layer actually has to express.
- policy engine
- agent safety
- audit
The homeostat: why autonomy needs a set point
Autonomy without a set point is drift. Ashby's homeostat is the missing piece in most agent architectures, and the one that makes them safe to leave running.
- homeostat
- cybernetics
- autonomy
One data model beats eleven integrations
Integration is the tax you pay for having split the data in the first place. What changes when orders, stock, cash and people live in a single model.
- data model
- integration
- ERP
One kernel for agents, robots and devices
A policy written for the server should hold on the warehouse floor and in someone's pocket. Why software agents and embodied ones belong on the same OS.
- robotics
- edge
- identity