Lindsay Edwards

Give The Robot A Spending Cap

I like handing work to AI agents and walking away. What I do not like is the version of that where I wake up to a bill instead of a feature.

This week in ContentForge I added a path for users to plug in their own AI provider key (Gemini, Claude, OpenAI) once they pass what their plan covers. The box for the key was the easy part. The part that matters is what sits behind every AI call: a layer that counts each call, enforces a daily cap and a monthly per-feature budget, and pauses an agent the moment it hits the ceiling.

Here is why that earns its keep. An AI feature is a tap you have left running. Every generated draft, every summary, every retry costs real money somewhere, and an agent stuck in a loop drinks faster than any person would. Without a cap, one bad prompt at 2am is a phone bill with no plan attached.

The take. As more of a product becomes AI doing the work, “how much did that cost” stops being an afterthought and becomes a feature you build on purpose. You would not give a new hire a company card with no limit and no statement. Same rule for the agents. Let them work, meter every call, and keep a switch that says stop.

Cheap to run is a decision you design in, not a number you hope for.

Keep reading