Fully autonomous AI demands a new enterprise anchor

Fully autonomous AI has crossed the threshold from concept to commercial reality, forcing enterprises to rethink the very foundation on which they run their businesses.

The shift is arriving faster than most organizations anticipated, bringing with it an urgent new set of economic and architectural demands. But the pressure is especially acute for enterprises that still treat legacy infrastructure as the center of their technology universe, according to John Roese (pictured), global chief technology officer and chief AI officer at Dell Technologies Inc. Now, as agentic AI moves from buzzword to board-level priority, the conversation has gone from possibility to implementation — and cost.

“I was introducing people to the word ‘agentic’ about a year ago,” Roese said. “Now we’ve realized that this idea of fully autonomous AI systems — of really shifting work into the machine layer — is now very real. At the same time … the word ‘tokennomics’ is now in our vernacular because we’ve realized that when you put these things into production at scale, the cost of different patterns of what you deploy are incredibly variable. Having a smart, intelligent way to use your hybrid infrastructure, to put the right workload in the right place to use the right model, is actually not just a nice-to-have — it’s required.”

Roese spoke with theCUBE’s John Furrier and Dave Vellante at Dell Technologies World 2026, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed fully autonomous AI infrastructure, the economics of hybrid deployment and the emergence of AI-centric engineering as a new organizational discipline. (* Disclosure below.)

Fully autonomous AI demands a new enterprise anchor point

Roese has consistently suggested that success in this era won’t come from chasing every AI breakthrough, but from building infrastructure capable of keeping pace — a view that puts the company squarely at odds with how most enterprises are currently approaching the problem. The legacy-versus-AI debate is one enterprises are increasingly getting wrong, with many attempting to bolt AI capabilities onto existing legacy, or “brownfield,” environments, he added.

“The biggest impediment to navigating the legacy environment into the AI era is this mistake of thinking you’re doing it by having the legacy environment be the anchor,” he said. “The only way it works is [if] you actually have an AI environment. You have to have a place to build agents and run them autonomously. You have to have a security architecture that works with agentic. The brownfield becomes a feeder system to the new AI environment, but if you don’t have the AI environment at all, you’re just putting some AI features around a brownfield.”

That infrastructure philosophy extends directly into data. The Dell AI Data Platform is an example of what it means to treat data not as a separate concern but as an integral component of the AI factory itself, according to Roese. The hardest work in any advanced agent deployment has been data plumbing, specifically converting institutional knowledge into knowledge graphs and ontologies that agents can actually use.

“An agent that just has access to a large language model is not an enterprise agent,” he said. “An enterprise agent also has access to a knowledge graph that organizes the proprietary enterprise knowledge. If you give an agent both of those, and the knowledge graph is ground truth, it behaves better, it is more relevant, it is more aligned to your business.”

As enterprises mature those agentic deployments, the question shifts from whether agents can do the work to whether they can do it reliably. Rather than allowing agents to improvise, Dell has embedded digitized process knowledge directly into its agentic systems — giving them a defined playbook to draw from, Roese explained. The distinction matters because unpredictable agents, however capable, cannot be trusted with the workflows that actually run businesses.

“We are big believers in digitized process,” he said. “In the old world, that digitized process led humans around. In the new world, when we build agents, one of the tools they use to accomplish work is they go to our digitized processes. The value of giving agents more information about what the process is — how they should work, rather than letting them invent — just makes them more predictable and more scalable.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of Dell Technologies World 2026:

(* Disclosure: TheCUBE is a paid media partner for Dell Technologies World 2026. Sponsors of theCUBE’s event coverage do not have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE