Agent operations

Running real work through agents and measuring what happens.

A small company runs its daily operations through a crew of software agents, each with its own name and job, working over plain text files. Autonomy is declared per project and enforced in code; every change is supposed to be reviewed by a session that never saw the author's conversation. The system is observable end to end. That makes it an instrument: a record of what agents actually do when the work is real, checked against what happened rather than taken on trust. It is aimed at a company that runs most of itself, with the failures still legible on the days it doesn't.