Every AI Company Is Accidentally Building a Bank — Dor Sasson, Stigg

This post argues that AI companies are facing pricing emergencies due to an underlying infrastructure problem: usage is checked after invoices are generated, not before. It proposes a fundamental shift towards synchronous entitlement checks before inference and asynchronous settlement, akin to ATM operations. The talk details the banking-like financial mechanisms required for AI products, including hold-and-settle, double-entry bookkeeping, credit sources, and hierarchical spend caps for teams, users, and agents. It outlines four distinct patterns for selling AI at scale, from reserve-and-settle to granular spend visibility, framing these as essential infrastructure for sustainable AI monetization.