200 agents, 2,011,438 tool calls: who's paying for your AI?

A recent analysis of an agentic system handling over two million tool calls reveals that the majority of AI costs are driven by inefficient model usage rather than complex reasoning. The author argues that most sub-tasks, such as data formatting or tool classification, do not require expensive frontier models. Instead, teams are often overspending by using a single 'default' model for every step of an agent's loop. By implementing a routing layer that assigns sub-tasks to the most cost-effective model capable of performing the specific action, organizations can significantly reduce expenses. Furthermore, this routing approach allows for the integration of compliance and data residency requirements directly into the workflow. The article concludes that most high AI bills are essentially 'routing bills' in disguise, and suggests that developers should map every sub-task to the cheapest model that meets its quality requirements to optimize budgets.