Confidence Is Not a Signal
The essay's core finding is that interviewers mistake their own certainty for validity. Charisma, likeability, and a shared alma mater flood the signal. What actually predicts performance — work samples, structured questions, base rates — gets crowded out. A founder who spends two years tuning evaluation harnesses for agents and zero minutes structuring human hiring has a strange blind spot: rigorous evals for the machines, vibes for the people.
The Humans Who Stay Must Be Worth Keeping
An AI-native company still needs humans at the consequential tip of the pyramid: goals, values, irreversible calls, anything you would defend in a room with a customer. That makes the cost of a bad human hire higher, not lower. When agents own the broad, reversible base of the work, every human seat is a judgment seat. Selecting for judgment with an unstructured conversation is like shipping an agent with no eval and hoping for the best.
Designing Trust Into the Selection Process
The essay's fix is structural: standardize the questions, score before you discuss, use a work sample that mirrors actual stakes. This is approval-layer thinking applied to hiring. Match the rigor of your selection process to the stakes of the role, the same way you match human-in-the-loop checkpoints to the stakes of an agent decision. A typo fix and a wire transfer should not share a workflow; a junior contractor and a judgment-seat hire should not share an interview format.
Base Rates Are a Founder Discipline
The essay returns repeatedly to base rates — what does this role actually predict, across all candidates, not just the one you liked this morning. Founders are base-rate blind by temperament; conviction is the job. But conviction about people, unanchored to structured evidence, produces the same failure mode as deploying an agent without a feedback loop: confident, fast, and wrong in ways you only see later.
