A hard budget cap is a feature of a pay-by-usage service or API that lets a customer say “after $X/month, cut this thing off and return errors”: once the configured spending limit is reached, the service stops rather than continuing to accrue charges. It is defined against a soft cap, which only sends a warning — for example an email — when the threshold is crossed and lets usage continue. Simon Willison uses the phrase default hard budget caps for his proposal that such a cap be on by default, with removing it an explicit opt-in.
Usage
Willison frames the need in terms of AI Coding Agent use, in We’re going to need default hard budget caps on pretty much everything: coding agents and personal agents greatly reduce the friction of spinning up code that does useful things, and some of that code costs money — calls to paid APIs, hosted web applications, or systems that bill for additional storage and compute. A soft cap does not protect against the case he describes, a rogue service consuming hundreds or thousands of dollars of usage overnight while its owner sleeps after a midnight warning email. He proposes that opting out of the cap be a clear, prominent checkbox in which the user accepts that the application will not be shut down and that they are responsible for subsequent charges, and that agents themselves could favour recommending providers that offer hard caps and warn inexperienced builders away from uncapped services.
He points to two cloud providers that had recently added features of this kind: AWS introduced a monthly spend limit for projects that pauses a project for the month once its limit is reached, still rolling out to a limited number of customers, and Google Cloud introduced Spend Caps, which set a monthly financial cap on specific services within a project.
When It Applies
The proposal is aimed at pay-by-usage services where runaway usage can produce an open-ended bill. Its main trade-off is stated by Willison himself: businesses may not want hosted applications to start returning errors because a budget was exceeded. His answer — that most businesses and individuals would prefer errors to a surprise bill of $10,000 or more — is his own expectation rather than a measured finding, and making the cap the default is his proposal in a single opinion post rather than an established convention.