Agentic AI by Dell: Transforming Work and Security – Discover How

The conversation at The Six Five Summit wasn’t just about faster chips or bigger models. It was about a fundamental redesign of what work is. John Roese, Dell Technologies’ Global CTO and Chief AI Officer, cut through the hype with a simple, powerful statement: agentic artificial intelligence isn’t just another tool; it’s about “digitizing work.” You are literally shifting work from a human being to a machine. This shift, moving beyond chatbots to autonomous systems that perform entire workflows, is forcing a complete re-evaluation of enterprise infrastructure, economics, and security. It’s not an incremental upgrade; it’s a structural change.

For the past year, Dell has been running this experiment on itself. Roese shared that the company moved from developing agents to putting them into production, handling tasks from scrubbing CRM data to managing special pricing approvals. The initial lesson was one of ruthless focus. Dell started with 900 AI projects and canceled most, whittling the list down to about thirteen strategically chosen deployments. The goal wasn’t just efficiency for its own sake, but to fundamentally “decouple revenue growth from its cost structure.” This is the agentic promise: automating categories of work, not just tasks, to change the financial equation of a business.

But here’s where the rubber meets the road, and the infrastructure demands get real. Matt Murphy, CEO of Marvell Technology, provided a crucial technical reality check. The infrastructure that excels at training massive AI models is not the same as what’s needed for production-scale agentic AI. Training is a compute-heavy sprint. Agentic inference is a memory-intensive marathon. Agents retain context, manage state across long chains of reasoning, and coordinate with other agents. A single workflow might involve twenty or thirty “turns,” each consuming memory. Murphy pointed out that in these scenarios, you’ll often run out of memory bandwidth long before you tax the compute capacity. Furthermore, the classic “throw more GPUs at it” solution hits a wall with latency. When you have hundreds of agents working in concert, the coordination overhead creates tail-latency issues that can’t be solved by simply adding nodes. The orchestration role of the CPU becomes critically important, managing the complex branching logic and tool calls that define agentic behavior.

This technical evolution drives the most profound change: the reshaping of jobs themselves. Roese was adamant about moving past the simplistic “AI takes your job” narrative. Instead, he frames a job as a container holding different types of work—productivity tasks, routine hygiene activities, coordination, expert knowledge and human-facing interaction. Agents are designed to extract specific slices from that container. “They do not take your job, they change your job,” he stated. Dell’s analysis of 6,800 roles led to a striking conclusion: every single job will change because every job contains work that can be digitized. The engineer freed from writing boilerplate code by an advanced coding agent can focus more on system architecture and customer requirements. The outcome is not elimination, but evolution toward more expert and human-centric responsibilities.

However, managing this evolution requires a new operational and security playbook. You cannot manage a fleet of autonomous agents with the same governance used for departmental software. Roese emphasized that success requires deliberate, top-down organizational change focused on outcomes, not just isolated automation projects. Economically, a one-size-fits-all approach to AI infrastructure is a dead end. The “token economics” for an agent supporting a high-stakes executive decision are entirely different from those for an agent cleaning data entries. Consequently, Dell itself operates a hybrid AI architecture, blending open models on-premises, frontier models in its data centers, controlled virtual private clouds, and device-based processing. This mix is necessary, Roese argued, because “you cannot do that with a monoculture” when enterprise work is so diverse.

The most urgent wake-up call, however, was on security. The old perimeter-based models are obsolete. Dell’s new requirement is stark: any autonomous agent—internal or external—that accesses Dell data must carry a Dell-issued digital identity. This isn’t just about authentication; it’s about creating a fine-grained authorization system with a built-in kill switch. It’s a recognition that “headless agents” operating independently represent a new attack surface. The security model must evolve from protecting user accounts to governing machine identities capable of autonomous action. While threats like post-quantum cryptography loom, the immediate challenge is building an identity and governance framework for a workforce that is increasingly non-human.

The message from the summit was clear. The era of agentic AI marks a transition from using AI as a powerful assistant to building it into the operational fabric of the enterprise. It demands hybrid infrastructure, reshapes roles and requires a completely new security paradigm based on machine identity. Companies that view this as merely a more advanced chatbot will be left behind. Those that see it as an opportunity to digitize work itself—to thoughtfully redesign what gets done and by whom—are laying the foundation for the next decade. The work is already shifting. The question is whether our infrastructure, our organizations and our security models can keep pace.

• Agentic AI isn’t just another tool
• Shifts work from humans to machines
• Requires a new security paradigm
• Demands hybrid infrastructure
• Reshapes job roles
• Focuses on outcomes over isolated projects