The AI Vampire

A confession staged in the middle of a victory lap. After a year of telling everyone that agentic coding works, Steve stops to count what it costs the humans running it — by his own admission "an unusually hard post to write, because it flies in the face of everything else going on." The vampire is borrowed from What We Do in the Shadows, and the symptoms are physical: nap attacks after long agent sessions, colleagues pricing nap pods for the office, AI-native startups running their people like a gold rush on "the same six tired pitches." At the center is a short thought experiment about who keeps the value when one employee quietly becomes 10x productive; both of its endings are bad. His way out runs through a formula he first chalked on an Amazon whiteboard around 2001, back when a different company was working its people like sled dogs, and arrives at a specific, startlingly small number for how long the new workday ought to be.

Steve doesn't exempt himself. Forty years of experience, unlimited time and tokens, the dial "set to 11" while he leans on it trying for 12: an unrepresentative outlier, by his own accounting, whose pace employers now squint at and mistake for a baseline. That candor is also the essay's soft spot, since a plea for moderation from someone who admits he can't practice it invites exactly the pushback it got — the replies collected below spent the following weeks arguing whether the vampire is a culture problem or ordinary economics wearing a cape. He typed it at the mall, he says, minutes from closing the laptop and going for a walk.