As AI agents take on enterprise tasks, companies face a new battle over access and control

Enterprise private AI has progressed beyond pilot projects, revealing a challenge surrounding control and access to agents executing tasks such as coding, claim processing, and workflow management. These agents require models, tools, and data to function effectively, yet many organizations hesitate to deploy autonomous software without oversight.

Consequently, platform teams are once again taking charge of enterprise architecture, employing orchestration, telemetry, observability, and role-based access control to manage fleets of agents. Purnima Padmanabhan, general manager of Broadcom Inc.'s Tanzu Division, explained that "agents have agency," meaning they interpret assigned tasks and act accordingly.

To balance speed, security, and sandboxed execution, Broadcom introduced the Tanzu Platform Agent Foundations, an integral component of VMware Private AI Cloud. This runtime ensures every model, tool, and dataset is explicitly linked, while curated data products manage chunking, vectorization, and access control. Padmanabhan emphasized that "the right way to secure an agent is to put it in a black box and give it nothing, but then you won't get any intelligence."

However, trust in the underlying code remains crucial, as agents are built with the provided libraries. Broadcom is scanning both commercial repositories and open source code for potential vulnerabilities, with June's Spring patch release being the largest in 23 years. For business leaders, the decision to adopt private AI hinges on timing, as agents can yield measurable gains and workflow automation, positioning organizations for competitive advantage.