You Can't Trust Your AI Bill. And Neither Can I.

Cost, chaos, and what engineering teams get wrong about building on AI

You've been told tokens are getting cheaper. You've been told AI makes engineers 10x more productive. You've been told the cost curve only goes down. What nobody told you: tokens are 1,000x cheaper than they were three years ago — and your aggregate spend is up 320%. William Stanley Jevons called this in 1865 watching coal. We're calling it Tuesday.

This isn't a FinOps talk. It's a reckoning.

In this session, CloudZero founder Erik Peterson walks through the structural reasons why AI billing is fundamentally different from cloud billing — hidden reasoning tokens, tokenizer gaps, the MCP overhead nobody measured, and the o1-vs-o3 pricing paradox that nobody has explained. He'll replay the real postmortems: the $47K eleven-day loop, the $82K stolen key, the five-figure eval bill, the engineering team whose annual AI budget was gone in weeks.

Then he makes the case for a new set of engineering-first metrics and architectural patterns — economic constraints as first-class design, circuit breakers for agents, model routing by margin — that let you build AI systems that survive production without destroying your P&L.

The central question: every engineering decision is a buying decision. Was yours worth it?