Autonomous agents don't use inference the way a human-prompted app does. That difference alone can bankrupt a product.
Ilan Zerbib (Founder & CEO,
@sapiom) made the point on the latest episode of Tokenized with hosts
@sytaylor and
@bamazizimesh. A human decides when to make each request. An agent can call the model whenever it wants, with no one pacing the spend.
"The cost of inference grew from like zero, spiked to $1.2 million after three months of operation."
🎙️ Listen to the latest episode on