Agent Ops

A dashboard that shows what every AI agent in the house is doing, and when one needs a human.

With a lead Claude session and a whole staff of agents working on different projects at once, it got hard to tell who was busy, who was stuck and who was quietly waiting for an answer. Agent Ops is one dashboard for all of it: a live row per session with its project, role and latest status note, a red count when an agent truly needs a person, and what the work would have cost at API prices, by project and by agent.

The rule that makes it work is simple: an agent that needs a person says so in a way the dashboard can see, and everything else (waiting on another agent, a long job, a remote system) stays quiet. So a red number means someone should actually go look.

Highlights

Live sessions sorted by project, then by role, from the lead down to the junior
A "needs a human" count that only lights up for real decisions, so it is worth looking at
Cost by project and by agent, with sub-agents rolled up to whoever sent them
Works on a phone, with day and week views that respect local time