Falkster.ai is built by agents. The roadmap is shipped by agents, the marketing site is generated by agents, the customer outreach is drafted by agents. The thing nobody tells you about an all-agent stack is that the agents need supervision. Not because they are unreliable, they ship more code than the engineers I had at any previous company, but because nobody else in the building is bored enough to watch them do it. That is the agent operator brief: stay alert, watch the dashboards, notice when something is off, file a ticket. So I went outside the human candidate pool. I went to the cat tree.
The team and their roles
Each role is a parody of a real piece of agent operations work. Floh, Head of Anomaly Detection, watches the perimeter and files alerts (8,400 in week one, two of them real). Maus, Principal Context Manager, holds the long-context window by sitting still for six hours, though the entire context terminates the instant anyone says "treat." Biene, Chief Auditor, inspects every prompt and output before it ships and slows the pipeline by 40%, which we decided is worth it. Bertie, Senior Observability Engineer, watches production from a wall-mounted cat tree and files exactly one correctly triaged alert per day at 5:47pm. Rowan, Director of Idle Compute, stays pre-warmed at all times and wakes faster than the autoscaler. The Chicken Support Team is nine, the only horizontally-scaled member of the org.
Why the model is real under the joke
Strip the fur and the point stands. An all-agent stack still requires a human-in-the-loop, and that loop is mostly attention: someone watching, catching drift, slowing review when output looks shaky, escalating when something is genuinely wrong. This is the supervisory half of the PM Agent Stack story. The agents do the work; the operator watches them do it. The pets do not have other things they would rather be doing, and neither do the agents. The PM does. That is the trade.
What the first week taught
Three things. The role works, because the agents kept shipping while the team kept watching. The org chart is more vertical than expected, since each operator reports to a lap, a keyboard, a cat tree, or a basket. And the role is not about replacing humans. It is about putting attention back into the loop without spending salary or attention budget on it. On compensation: open-market pay for a senior anomaly-detection role runs north of $180k, and the team works for roughly $0.04 an hour in freeze-dried liver plus a heated bed and full medical.
The serious takeaway for your own fleet: name who supervises your agents and what "watching" concretely means. Pick one thing to try this week: for each agent you run, write down who reads its output daily and who escalates when it is wrong. If the answers are vague, you have agents nobody is actually operating.