The AI agent bottleneck isn't model performance — it's permissions

Workday argues that enterprise AI agents are stalling because companies lack reliable permissioning and governance, not because models are too weak. Its Sana system uses Workday’s HR and finance records, security models, approvals, and Gemini-based reasoning to ensure agents act accurately and only within authorized boundaries.

The article highlights a shift from model performance to agent governance as the critical enterprise AI challenge. Systems of record like Workday may become control planes for AI agents, especially in high-stakes workflows where errors in pay, scheduling, or finance can cause immediate harm.