Making AI Headlines · Page 6
Garry Tan's Agents Work Late, and Their Token Bill Keeps Pace
· New York time
Y Combinator's Garry Tan says he spends up to $200,000 a year on AI tokens as he works with agents. Speaking on Walter Isaacson's podcast The Builders, Tan also said he sometimes barely sleeps because he is excited to use them. The figure is a vivid illustration of the cost of intensive AI use. Tokens are a unit used in processing AI requests, so repeated work can translate into a substantial bill. Tan's upper-end figure is his account of his own spending, not a typical subscription price or a forecast for other users. Agents can make it tempting to keep adding tasks, because a system can continue processing requests when its human operator would ordinarily be done for the day. The financial question is how much value that activity produces relative to its cost. A powerful workflow is not automatically a cheap one, particularly when enthusiasm keeps the meter running. The old promise of automation was that the machine would do the work while the human rested. Tan's account offers a more modern variation: the machine does the work, the human stays awake watching, and both contribute to the invoice. The agent may never need coffee, but the person checking its progress could still have a very long night.
Read this story in the full newspaper →Satire. Not factual reporting. Read the original reporting from the linked publishers.