Two agents got stuck talking to each other. It took eleven days and $47,000 for anyone to notice.
No alert fired. No budget control tripped. They executed perfectly inside their parameters, which is why nothing caught them.
Every call your agents make is a financial decision, executed in milliseconds without approval, and most enterprises won't see the total until the invoice closes. That gap is Agent Debt. It's structural: observability, API security, and FinOps each answer a different question, and none of them is whether the spend should happen at all.
What you'll find inside:
- The e-commerce team whose agent costs went from $5,000 to $50,000 a month between prototyping and staging, and the three behaviors behind it
- Moffatt v. Air Canada: the tribunal that rejected "we aren't responsible for what our AI says," and what that means for autonomous spending
- Why Gartner expects 40% of agentic AI projects canceled by 2027, and why all three reasons they cite are governance failures, not capability failures
- A six-layer framework, including the two layers that can deny, downgrade, or reroute a call before the money moves