I wasted $47 on AI APIs because I had no spending control

Yesterday I was testing an AI agent.
It was calling APIs automatically.
Everything was fine… until I checked my usage.
I had wasted $47.
No limits.
No tracking.
No control.
And I realized something:
AI agents today can:
think
act
call APIs
But they can’t manage money
Which makes them dangerous at scale.
So I started exploring a solution: https://agent-purse.vercel.app⁠�
Still early, but I’m curious:
How do you control spending in your AI agents?