Why AI Agent Permissions, Not Model Speed, Stall Enterprise Automation

Enterprise AI agents are hitting a wall, not because of model accuracy, but due to AI agent permissions.

AI Agent Permissions: The Real Bottleneck

Every workflow that hands a task to an autonomous bot eventually asks: what data can it touch, whose authority backs the action, and how does the platform verify compliance? Workday’s answer is to embed its legacy system of record as the governing layer for its Sana agents. Gerrit Kazmaier, president of product and technology, told Reuters that customers “struggle when they cobble together solutions for their agents.” “Sana makes sure the integrity of the approvals and security model is always adhered to,” he said.

“Frankly, that’s where we see customers struggling when they try to build do‑it‑yourself AI by just accessing raw data, so the richness of the security model gets lost, and the results become overly broad.”

Workday launched Sana in March and swiftly expanded a partnership with Google, feeding the Sana system of record into the Gemini Enterprise so agents are discoverable across both platforms. Accuracy in HR and finance is non‑negotiable; a single mis‑pay or mis‑schedule can cascade into regulatory exposure. Kazmaier noted that “almost right is not acceptable” when payroll, ledger closing, or shift planning are at stake. To mitigate risk, Workday built Gemini as the base reasoning engine, layered with a context engine and business‑process logic, then added verification and classification models that “interrogate” outputs before execution. The question of accuracy collapses into identity: does the system know enough about the agent, the authorizing human, and the current state of the record? Workday can infer organizational hierarchies from client data; third‑party identity providers such as Okta already cross‑check Workday, making it the de‑facto system of record for many enterprises. The Sana Self‑Service Agent uses Gemini as a conversational front‑end; the user is authenticated through Workday’s identity model, and the agent can act only within that user’s permissions. Audit trails follow the same logic: Gemini logs interactions, while the authoritative audit resides in Workday. For regulated domains, the governance layer must live inside the system of record, not beside it. Dan Obendorfer of Würk told Bloomberg that “if your permissions are defined somewhere outside of where the data actually lives, you’ve already lost.” Kadan Stadelmann, CTO of Compance.AI, warned that “without agent ownership, performance, costs or actions, chaos ensues.” Workday reports a ↑ 12% rise in enterprise agent deployments since the Gemini integration, suggesting that tighter permissioning may finally unlock the promised productivity gains.

Intel provided by: Dr. Aris Thorne

Artificial Intelligence Researcher