Enterprise private AI has moved past the pilot stage, and the shift is exposing an awkward gap. Agents that write code, process claims and run business workflows need models, tools and data to be useful, yet few organizations want autonomous software wandering across their infrastructure unsupervised.
That tension is putting platform teams back at the center of enterprise architecture. The same cloud-native disciplines — orchestration, telemetry, observability and role-based access control — are now being applied to fleets of agents. Because agentic applications are essentially microservices applications, the platform is a natural place to enforce control, according to Purnima Padmanabhan (pictured), general manager of the Tanzu Division at Broadcom Inc.
“Agents have, by definition, agency, which means you just give the intent and resources and then the agent interprets that intent and decides to do something,” Padmanabhan said. “You have to say, ‘OK, I want to be able to build agents fast, I want to be able to build that securely and I want to run them, but I want to run them in a sandboxed way.'”
Padmanabhan spoke with theCUBE’s John Furrier at VMware Explore 2026, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed agent runtimes, trusted open source and the data foundations behind production AI. (* Disclosure below.)
Private AI gets a deny-by-default agent runtime
Broadcom’s answer to the challenge of securing autonomous agents is a runtime rather than a toolkit. Tanzu Platform Agent Foundations is now an integral part of VMware Private AI Cloud, giving agents a sandboxed place to run where every model, tool and dataset must be explicitly bound, Padmanabhan explained. Curated data products sit alongside it, handling chunking, vectorization and access control so agents can work with curated data without directly accessing original sources.
“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,” Padmanabhan said. “I want to give it on my terms in a curated way with identity, with [role-based access control], with credential management. Being able to connect it to curated data sets all on a single platform means I can move faster.”
Trust in the underlying code matters just as much, since agents build with whatever libraries they are handed. Broadcom is scanning both its commercial repositories and open source with Mythos and expanding its Spring and Java security work into Python and Node.js, with June’s patch release the largest in 23 years of Spring stewardship, Padmanabhan noted.
“We are finding issues. We are finding vulnerabilities,” she said. “These vulnerabilities are not just your low-level vulnerabilities, these are things that can completely bring down an enterprise. We take that job very seriously.”
For business leaders, the calculation is less about technology than timing, as coding agents deliver measurable gains and workflow automation follows close behind. Private AI, in that framing, is infrastructure for a competitive position rather than a science project. Organizations that move quickly can turn AI into a source of revenue and competitive advantage, while those that wait risk falling behind, Padmanabhan noted.
“I think the message to the CIO is this is the chance to shine, right? Because this is a revenue opportunity for business,” Padmanabhan said. “It’s also a competitive advantage, and if you don’t do it, it’s a competitive disadvantage, because everybody else is going to be doing it.”
Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of VMware Explore 2026:
(* Disclosure: TheCUBE is a paid media partner for the VMware Explore 2026 event. Neither Broadcom, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)
