The real bottleneck: permissions, not model performance

Enterprise AI agents are hitting a wall, and it is not because the underlying models lack intelligence. According to a recent report from VentureBeat, the primary obstacle is permissioning — determining what data an agent can access, on whose behalf, and how the system enforces those boundaries. Every agentic workflow eventually confronts the same question: what is this agent allowed to touch? Without a clear answer, deployments stall. Gerrit Kazmaier, president for product and technology at Workday, told VentureBeat that customers often struggle when they cobble together solutions from disparate tools, leading to security gaps and compliance risks. The issue is not unique to any single industry; it affects finance, healthcare, and retail organizations alike. As companies rush to deploy AI agents for tasks like customer support or internal data retrieval, the lack of a unified permission model becomes a critical failure point.

Workday's solution: system of record as governance layer

Workday is proposing a straightforward fix: use its existing system of record as the governance layer for AI agents. Kazmaier explained in the VentureBeat interview that Workday already manages employee data, roles, and permissions for human users. Extending that framework to AI agents is a natural evolution. The system would define what each agent can access based on the same role-based access controls used for employees. For example, an agent tasked with retrieving payroll data would only see records for which the requesting manager has authorization. This approach eliminates the need for separate permission models, reducing complexity and potential errors. Workday claims this integration can cut deployment time for agentic workflows by up to 40%, according to internal estimates shared with VentureBeat. The company is testing this with a handful of enterprise customers, with a broader rollout expected later in 2026.

Competitive landscape: who else is tackling agent permissions?

Workday is not the only company addressing the permissions bottleneck. Several startups and established vendors are developing solutions to govern AI agent access. For instance, Okta announced in early 2026 a new module for agent identity management, allowing organizations to assign temporary credentials to AI agents. Similarly, Salesforce has integrated permission controls into its Agentforce platform, though analysts note it relies heavily on manual configuration. Microsoft has taken a different approach with its Copilot extensions, using graph-based permissions from Azure Active Directory. However, Kazmaier argues that these solutions often lack the contextual depth of a system of record, which knows not just who an employee is but also their role, department, and reporting structure. This context is critical for nuanced decisions, such as whether a finance agent can access budget data for a project the user does not manage. The market is still fragmented, with no clear leader, but Workday's tight integration with HR and financial data gives it a unique advantage.

Implications for enterprise adoption and security

The permissions bottleneck has broader implications for enterprise AI adoption. Without robust governance, organizations risk exposing sensitive data to unauthorized agents, leading to compliance violations under regulations like GDPR and HIPAA. A 2025 survey by Gartner found that 62% of enterprises delayed AI agent deployments due to security concerns, with permissioning cited as the top issue. Workday's approach could accelerate adoption by providing a familiar framework — the same role-based access used for human employees. This reduces the learning curve for IT teams and aligns with existing audit trails. However, critics warn that relying on a single vendor's system of record could create lock-in, especially for organizations using multiple HR and finance platforms. Workday counters that its open API strategy allows integration with other identity providers, though the governance logic remains proprietary. For now, the company is betting that simplicity and security will win over enterprises wary of complex multi-vendor solutions.

Future outlook: agentic workflows beyond 2026

Looking ahead, the permissions challenge will only intensify as agents become more autonomous. Workday envisions a future where agents not only retrieve data but also execute transactions, such as updating employee records or initiating expense approvals. Each action requires granular permission checks, potentially in real time. Kazmaier told VentureBeat that Workday is exploring dynamic permissioning, where an agent's access rights adjust based on context, such as the time of day or the sensitivity of the request. This would require machine learning models to predict risk levels, adding another layer of complexity. Industry observers expect that within two years, most enterprise software will include built-in agent governance features, making standalone permission tools obsolete. Workday's early move positions it as a potential standard-setter, but competitors are likely to follow quickly. The key will be balancing security with usability, ensuring that agents can act swiftly without compromising data integrity.