Hard budget caps should be the default for pay-by-usage AI services, argues Simon Willison

Simon Willison argues that pay-by-usage AI services and APIs should default to hard budget caps—limits that shut off usage and return errors once a configured monthly threshold is reached. Soft caps, which only send warnings, are insufficient because they allow costs to accumulate while users sleep or are offline.

Willison highlights the rise of coding agents and personal agents, which lower the barrier to deploying AI-powered applications that may incur usage-based charges for APIs, storage, or compute. Without hard caps, a misconfigured or rogue agent could generate hundreds or thousands of dollars in charges before a warning is noticed.

He contends that most users—whether individuals or businesses—would prefer an error to a surprise bill of $10,000 or more. Therefore, hard budget caps should be the default, with an opt-in mechanism for those willing to accept the risk of uncapped usage.

AWS introduced spending limits in mid-September, allowing users to set a monthly spend limit for a project; if the limit is reached, the project is paused for the month. Google Cloud launched a similar feature in July called Spend Caps, which lets users set a monthly financial cap on specific services within a project.

Willison calls on providers to make hard budget caps the standard and suggests that agents could help by recommending providers with such safeguards and warning users away from uncapped services that pose financial risk.