When your agents are doing execution, the humans you keep had better be excellent at the one thing agents cannot do: recognizing novel failure before it kills you. That makes the composition of your small human core more consequential, not less.
Vinod Khosla's gene pool engineering framework starts with a brutal prior step most founders skip: writing down every assumption that, if wrong, ends the company. Only after that list exists do you ask who belongs on the team — and the answer is whoever has personally retired that specific class of risk before, in a prior life, under real pressure. A résumé signals capability in the abstract; a gene pool slot signals that a particular failure mode has already been survived and understood by someone in the room.
When agents handle the execution layer, the risk register shifts in character — fewer risks about shipping speed, far more about trust collapse, model drift, and edge-case judgment under ambiguity. That means your gene pool analysis should now explicitly list those categories: who on your human core has navigated a trust breach in an automated system, who has managed regulatory exposure on opaque outputs, who has rebuilt confidence with a customer after an agent went sideways. The humans you are engineering into the team are not there to outpace the agents; they are there to have retired the failure modes the agents cannot see coming.
- Start every hiring conversation from the risk register, not the job description
- map each open role to a specific existential assumption your venture is still carrying
- in an agent-native company the most valuable gene pool slots are often the ones that look least like traditional operators.
