The case for hard spending caps by default

Simon Willison argues that any service billed by usage should ship with a hard budget cap turned on by default. The idea is simple: once spending passes a set limit for the month, the service stops and returns errors instead of quietly running up the bill. His worry is specific to this moment. Coding agents and personal agents now make it easy to deploy something to the cloud, often for people who have never had to think about how a pay-per-request backend can spiral. One runaway loop or a burst of traffic, and you wake up to a five-figure charge you never agreed to.

The pieces already exist, he points out. AWS added spending limits in September 2026 that pause a project when it hits its monthly cap, and Google Cloud shipped Spend Caps for individual services back in July. The catch is that these are opt-in and easy to miss, so the behavior you get without doing anything is still unlimited spending. Willison wants that flipped: caps on by default, with an explicit checkbox for the minority of users who knowingly accept open-ended charges. He keeps returning to one point in his argument: most people, businesses included, would rather get errors than a surprise bill. He even suggests agents should steer inexperienced developers toward providers that offer hard caps in the first place.

Why it matters

If you deploy anything an agent wrote for you, check now whether your provider can cut you off at a fixed dollar amount, because the default almost everywhere is still an unlimited bill, and the agent that generated your code will not be the one paying it.