We're going to need default hard budget caps on pretty much everything

Simon Willison wants a very specific kind of guardrail to become standard: hard budget caps that actually stop a service when it hits the limit. Not a gentle warning. Not an email that arrives after the damage is done. A real cutoff, with errors returned once the monthly spend is gone.

His case starts with coding agents and the less scary personal-agent versions of them. They make it far easier to spin up software that does useful work, which is great until that software starts calling paid APIs or running up costs on hosted apps, storage, or compute. At that point, a soft cap is basically a polite note attached to a mess.

He argues that most people would rather see an error than discover a surprise bill that runs into the thousands. For businesses, the objection is obvious: no one likes an app that suddenly stops because the budget ran out. But his answer is simple enough. If someone wants to accept that risk, fine — make it an opt-in choice with a clear checkbox that removes the cap and makes the customer responsible for whatever comes next.

Willison says AWS is the service he most wants this from, and he’s heard enough stories about runaway costs to understand why people avoid it for personal projects. AWS did finally launch spending limits a few weeks ago, with a monthly spend limit that pauses a project for the rest of the month once the limit is reached. The catch is that the feature is still only rolling out to a limited number of customers. Google Cloud added something similar in July, called Spend Caps, which can set a monthly financial cap on specific services within a project.

That makes this feel less like a niche request and more like an industry correction. If AI agents are going to keep making it easier to trigger real spend with very little friction, then the defaults need to be harsh. Friendly warnings are not a safety system; they’re a billing style.