AI BriefingThe Bottleneck for AI Agents Is Not Model Performance but Permissions
Key point
The bottleneck for enterprise AI agents arises not from model performance but from issues of data access permissions and governance.
Details
The reason enterprise AI agent adoption is being delayed is not a lack of model performance, but the issue of Permissioning, which determines what data an agent can access on whose behalf. If an agent accesses raw data directly without a security model, existing sophisticated security systems can collapse, risking outcomes that are excessively broad.
Workday addresses this problem through Sana, which uses Workday's existing systems as the agent's governance layer. Sana uses Google Gemini as its default reasoning layer, while combining it with Workday's context engine and business process logic on top to improve accuracy.
Especially in HR and finance, 'perfect accuracy' is essential rather than 'almost right.' Workday prevents errors by adding a model that validates and classifies the agent's output before execution.
Experts emphasize that permissions should be managed within the System of Record where the data actually resides. If permissions are defined outside of the data, ownership and control over the agent can be lost, leading to confusion.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.