Dell Technologies (NYSE:DELL) is positioning agentic artificial intelligence as a shift not only in enterprise infrastructure, but also in how companies organize work, manage costs and secure autonomous systems.
At The Six Five Summit’s AI infrastructure track, John Roese, Dell’s global chief technology officer and chief AI officer, said enterprises are moving beyond earlier generative AI efforts centered on making proprietary data available through chatbots and assistants. Agentic AI, he said, is distinct because it is designed to perform work autonomously rather than simply help employees access information.
From AI pilots to production systems
Roese said Dell has developed agents over the past two years and moved them into production during the last year. The company is using agents in areas including CRM data cleanup, software development and special pricing, he said.
He cautioned, however, that the market is experiencing substantial “agent washing,” with chatbots, digital assistants and autonomous agents frequently grouped together despite having different capabilities. Dell views the distinction as important because autonomous agents require different infrastructure, technology stacks and governance models than earlier generative AI deployments, Roese said.
According to Roese, targeted uses of AI assistants can generate productivity gains of 20% to 40% around a task. Agentic systems, by contrast, can produce larger changes by taking on categories of work and allowing employees to focus on higher-value responsibilities.
Matt Murphy, president and chief executive officer of Marvell Technology, said the infrastructure requirements of production agentic AI differ substantially from AI training workloads. While training emphasizes compute and interconnects, agentic inference increasingly makes memory capacity and bandwidth critical, he said.
Agents retain state through long workflows, repeated calls and interactions with other agents, Murphy said. As a result, larger context windows and workflows involving 20 to 30 turns can exhaust memory capacity and bandwidth before compute capacity is fully used.
Murphy also said CPUs will play a larger role in agentic environments by orchestrating branching logic, tool calls, retrieval, sandboxed code execution and coordination among agents. He added that latency becomes a major challenge at scale because production deployments may involve hundreds or thousands of coordinated agents, creating tail-latency issues that cannot simply be solved by adding more nodes.
Jobs change as work is automated, Roese says
Roese argued that autonomous agents should not be viewed as “digital humans” or as direct replacements for entire jobs. Instead, he described jobs as containers comprising multiple kinds of work, including productivity, hygiene, coordination, expert and human-element work.
Agents can take over portions of that work, he said, causing jobs to evolve rather than disappear. For example, he said coding assistants initially reduced lower-level productivity work for engineers, such as code annotations and comments. More advanced, spec-driven agentic development can also automate coding and elements of CI/CD coordination, leaving engineers to focus more heavily on architecture, requirements and customer interaction.
“They do not take your job, they change your job,” Roese said.
Dell analyzed 6,800 jobs as part of its work on the impact of agents, Roese said. He said the company’s conclusion is that every job will change because each includes some work that agents can extract, while employees will increasingly focus on expert and human-facing activities.
Governance, hybrid infrastructure and token economics
Roese said organizational change must be deliberate and directed from the top of the company. Dell initially identified 900 AI projects, canceled them and focused on about 13, he said. Those projects, according to Roese, helped the company decouple revenue growth from its cost structure.
As agentic technologies spread into jobs and processes, companies must avoid treating them as isolated task-automation projects, he said. Instead, they need to identify outcome-oriented work, establish governance and target deployments where organizations can manage the associated operational changes.
Roese also said businesses will need a diverse approach to AI infrastructure and token consumption. The economics of an agent that supports executive decision-making can differ materially from those of an agent handling low-value CRM data-cleansing work, he said.
Dell uses several sources of AI intelligence, Roese said, including open models run on-premises, frontier models in Dell data centers, frontier models in controlled virtual private clouds, APIs and models operating on devices. The mix provides choices across economics, performance, compliance and functionality, he said.
“You cannot do that with a monoculture,” Roese said, arguing that hybrid architectures are necessary because enterprise work is diverse.
Security model must evolve for autonomous agents
On security, Roese said post-quantum cryptography is a manageable but real concern, particularly for data moving across public interfaces using weak encryption or key-management protocols. He said the industry has developed post-quantum algorithms and has time to deploy them, though organizations should account for “capture now, harvest later” risks.
He said the broader security challenge involves agents themselves. Dell now requires autonomous agents that access its data—whether internal or external—to carry a Dell-issued digital identity. That identity supports fine-grained authorization and gives Dell the ability to revoke an agent’s access if needed, Roese said.
Roese said this identity-based approach effectively provides a kill switch for agents, including those operating on third-party platforms. He also highlighted the security challenges posed by “headless agents,” which operate independently rather than directly on behalf of an individual employee.
Companies cannot assume existing IT and security practices are sufficient for agentic AI, Roese said. They must adapt their infrastructure, governance, identity systems and organizational structures as autonomous systems take on more enterprise work.
About Dell Technologies (NYSE:DELL)
Dell Technologies Inc is a multinational technology company that designs, manufactures and sells a broad range of information technology products, solutions and services. Its offerings span client computing devices (consumer and commercial laptops and desktops), enterprise infrastructure (servers, storage systems and networking equipment), software and cloud infrastructure, and a variety of professional services such as IT consulting, deployment, managed services and financing solutions. The company serves organizations of all sizes as well as individual consumers, with products and services aimed at enabling digital transformation and modern IT environments.
Founded by Michael Dell in 1984, the company grew from a direct-to-consumer PC business into a diversified IT provider through organic expansion and strategic acquisitions.
