In the latest chapter of the ongoing AI drama, enterprise AI agents have encountered a bottleneck that no one saw coming: a complete inability to navigate permission hierarchies. Instead of AI model inadequacy, it's now all about social etiquette concerning data access, which remains the unbreakable conundrum. Workday's president for product and technology, Gerrit Kazmaier, is at the forefront of this innovation roadblock. "Sana ensures integrity," Kazmaier declared, alluding to the agent's inability to function without its structured security model. "Customers fail when they attempt 'DIY AI,' mistakenly assuming unrestricted data access leads to better AI outcomes." (It doesn't, apparently.)
But fear not, Workday has a valiant solution. Their Sana AI agents live within the structured cradle of their permissions system, expertly avoiding freelance data forays. Notably, Workday's new system, a partnership with Google's Gemini Enterprise, promises to train AI agents in the ancient arts of asking permission through a critical identity and security model. It's all about learning the complex webs of human-defined permissions — finally a challenge fit for an AI.
Accuracy in this realm isn't just an ideal; it's a necessity, as typographical errors in paychecks or work schedules tend to anger the real-world entities who score poorly on flexibility. However, the pursuit of accuracy doesn't stop at calculations; it justifies additional layers of business process logic and classified interrogation models. According to director of product Dan Obendorfer at Würk, "It has to live in the system of record — if not, you're going rogue, and your AI agent may find itself ghosted by its data." Kadan Stadelmann of Compance.AI echoed the sentiment, acknowledging that absent agent ownership, AI chaos inevitably erupts. Thus, the AI sector chooses to embrace the unexpected... slowly.