Recent research from VentureBeat shows that deploying AI agents in enterprise environments is hitting a major wall: access control and system governance, rather than the performance limitations of AI models.
Developments
Autonomous agent systems often stall when they hit these questions: What data is this agent allowed to access, on whose behalf, and how does the system verify that? Workday, a major player in enterprise solutions, recently introduced Sana—an agentic system that uses its existing 'system of record' as the governance layer.
Instead of allowing agents to indiscriminately access raw data, Workday integrates Gemini as the reasoning layer while keeping control and business process logic within its own security layer. This approach ensures accuracy for sensitive tasks in HR and finance, where even a minor error can have severe consequences.
Why It Matters
For Vietnamese enterprises looking to adopt Agentic Workflows, the challenge is not just choosing between GPT-4 or Claude 3, but how to integrate them into existing permission systems (such as IAM or LDAP). A "Do-It-Yourself" (DIY AI) approach that grants agents access to raw data can lead to data leaks or overly broad outputs, compromising system security. The trend of using the security layers of management software (like Workday or SAP) as a "brake" for AI agents is set to become the new standard.