If your agents are only as good as the humans steering them, you cannot afford the polite fiction that talent is fungible. The irreplaceable person problem just got more expensive.
Dan Luu's essay "Individuals Matter" dismantles the comfortable management belief that organizations are the unit of performance and people are interchangeable parts. He works through concrete historical and technical cases where a single engineer, researcher, or decision-maker changed outcomes by an order of magnitude — not because they were lucky, but because the distribution of capability is far wider than org charts acknowledge. The lesson is structural: if you design as though no one is irreplaceable, you build systems that lose irreplaceable people and never notice.
In an AI-native studio, agents handle volume and the humans handle judgment — which means the judgment layer is smaller, more concentrated, and higher-stakes than ever. The one person who can read a bad agent output and know exactly why it went wrong, or who built the mental model of your domain deeply enough to catch the failure nobody else would catch, is not a bottleneck to be designed around. Luu's argument translates directly: your hiring bar, your retention decisions, and your trust architecture all need to account for the real variance in that small group of humans, not the average.
- Treat the variance in human judgment as the primary risk variable, not an HR platitude
- build your oversight layer to surface who the load-bearing people actually are before they leave
- design agent workflows so that rare expertise gets used at the decision point, not buried in review queues.
