Most AI products today are one large chat window that does everything: code, research, images, calendar. The result is neither specialised nor easy to replace.
A better shape may be closer to a microkernel: each AI owns one job, isolated from the others with clear permissions, and a scheduler layer connects them. How context is passed, who may read my files, which step needs my confirmation — all defined in that layer.
The benefit is separability: modules can be built independently, swapped independently, tried in parallel. The prerequisite is isolation first — linkage without isolation only multiplies risk.
Life needs less overthinking. Do more. Feel more.


