Default hard budget caps needed for AI agent services, argues Simon Willison

Simon Willison, writing on his personal blog simonwillison.net on October 3, argues that pay-by-usage services and APIs need default hard budget caps, a feature that would cut off usage and return errors once a spending threshold is reached, rather than merely sending a warning email after the fact. Willison frames this as a response to the growing use of coding agents and personal agents, which he describes as coding agents operating with less human oversight.

According to the summary of his post, Willison distinguishes between “soft caps,” which notify a user that a limit has been crossed, and “hard caps,” which stop service delivery outright once the limit is reached. He states plainly that soft caps “will not cut it,” arguing the distinction matters because autonomous agents can generate usage and costs without a human monitoring each transaction in real time.

The post identifies this as a feature gap across “pretty much everything,” per its title, suggesting the issue extends broadly across pay-by-usage services and APIs rather than any single platform. The provided material does not name specific services, companies, or incidents that prompted the post, nor does it detail proposed implementation mechanisms beyond the hard-versus-soft cap distinction.

Willison’s post was published October 3, 2026, and is credited solely to him as author on his own site.