How to Build AI Evals for Tool-Calling Agents

Every other week it feels like a new model shows up with a shiny score on some "trust me bro"...