Salesforce VP: Cheaper AI tokens won't fix enterprise inefficiency

A Salesforce VP argues that enterprises are wasting 5-10 times more tokens than necessary due to architectural inefficiencies, not just model costs. The article identifies five key failure points in the AI pipeline: context bloat, ungoverned data access, using the wrong model, lack of persistent memory, and single-use semantics. It advocates for an architectural foundation that optimizes token usage through master data management, data governance, cost-routing models, structured memory, and caching. The piece concludes that token efficiency will become a competitive advantage as enterprises shift from maximalist AI adoption to cost-controlled, impactful outcomes.