‘There’s been tremendous agent washing’: Dell Technologies CTO John Roese says the real potential of AI agents is just being realized – and they could end up managing humans

AI agents could excel at managing workers and completing mid-tier tasks, according to first-hand testing from Dell Technologies.

John Roese, CTO and chief AI officer (CAIO) at Dell Technologies, told ITPro that the real value of agents is starting to emerge, even as unsophisticated tools are unhelpfully sold as ‘agentic’.

“There's tremendous agent washing,” Roese told ITPro.

“Everybody is calling everything they do an agent which is not true. Agents are autonomous systems, they are not a chatbot and we have been spending a great deal of time on actually maturing the ecosystem, getting real autonomous agents to work.”

Dell’s work on AI agents has included the development and delivery of the agent to agent (A2A) protocol and model context protocol (MCP), which help connect AI applications to external systems.

Dell defines an agent as a piece of software containing four distinct components:

• A large language model for ‘world knowledge’, including communication and reasoning.
• A knowledge graph, embedding specialized information and memory rules into the agent so it has access to data it wasn’t trained on and the capacity to build skills.
• MCP, an open standard for agent tool use and data access.
• Inter-working protocols such as A2A, which connect agents together in teams or fleets.

Roese contrasted this model with the approach of traditional and generative AI tools to date, which have focused on knowledge retrieval.

“Every tool to date has done one thing in AI: it's unlocked proprietary data,” Roese told ITPro. “That's what chatbots do, that’s what RAG does, ChatGPT does that, that's it, that's all it does – unlock data that exists somewhere that you can't access it.

“Agents do work, and you can't do work with a tool that was designed to just unlock data. You need an actual agent.”

AI agents could be managers one day

The first ‘wave’ of AI adoption was centered on unlocking this proprietary data, augmenting human capabilities with technology. This next chapter will go one step further, however.

“Agents aren't about that,” Roese said. “Agents are about the second part, they're about digitizing the skills, taking the skills that you have that are locked up in people's brains and making them scalable, making them repeatable with technology.”

The vast majority of deployed agents are intended as simple productivity tools or expert workers. But Roese said that recent testing by the tech giant found agents excel at completing simple tasks that aren’t being completed within organizations due to lack of time or low cost benefit.

“The big ‘aha’ moment was some of the work we can apply [agents] to is actually not done now by people, nor will it ever be done by people, because it's not worth doing it with people.

Indeed, Dell is beginning to deploy what it calls ‘steward agents’ or ‘hygiene agents’ and using them to clean its CRM data.

Roese gave the example of a private hospital in the UK which has struggled to keep its database on the specialties of doctors up to date. Historically, the only way to do this was to ask doctors to spend hours filling in forms, hire data workers to complete the task, or hire admin workers to do it at a poor return on investment.

Agents that work based on outcome, with access to data and reasoning capabilities, could constantly work in the background to keep this database accurate and current.

“You wouldn't say it's a high value task, but it's incredibly interesting if it happens because healthcare gets better but to do with people, or even augmented people, would be unaffordable.”

Much has been said about the potential for workers to become the ‘bosses’ of AI agent teams, with Salesforce CEO Marc Benioff having last year predicted today’s CEOs will be the last to oversee an all-human workforce.

But Roese suggested that the reverse will also be a trend in the near future, with agents used to coordinate complex human work.

“So what we started to try is, what if we give an agent the goal of making sure that the entire team accomplishes a task and we plug it into teams and communication streams?”

“It’s not the thing doing the work, but it knows that the goal is to build this product or to deliver this outcome, or to run this campaign, or whatever.”

For example, Roese suggested an editor within a news organization could use an agent to monitor the collaboration between staff writers on the team, with the system intervening to keep workers on track.

Dell has rolled out these coordinator agents in production environments and workspaces, Roese said, to overcome the “human entropy, the fact that large, complex processes rely on the goodwill of human beings to work well together”.

“So what we found is agents don't replace the humans but they act as this continuity manager, this coordinator,” he explained.

“Because they're on seven by 24, because they can reason, because they can do just a lot of work, forever, if you put an agent as the glue layer on a team of people and machines working together on a task, the entire task becomes more predictable.”

Throughout 2026, Roese added, firms will be surprised by the extent to which agents transform their organizations.

Dell’s AI ROI

In 2024, Roese told ITPro that Dell had narrowed its internal AI attention down from 800 potential projects to just four key areas that could deliver the most value: global supply chain, global services, global sales, and global engineering.

Since then it has deployed AI tools across these areas and is already seeing a significant return on investment.

“The net effect when we did our financial results after that full year of or just about a full year of those four big areas – in their first generation, it's just the starting point – was that our revenue last fiscal year grew by about $10 billion and our absolute cost declined by 4%.

“Now that that may not sound important, that's never happened in 41 years of Dell. Every other time that the revenue went up, the cost went up with it.”

This is not just a result of adopting AI tools, but transforming the entire company including its people and processes in anticipation of the “AI era”, Roese said.

For example, one of Dell’s major internal AI focuses in 2025 was to use AI to give back time to salespeople, allowing them to spend their day generating more money.

“It turns out 40% of their life was spent preparing to be in front of customers, that is now significantly less and the result is that they're just spending more time in front of customers,” he said.

At the same time, he added, Dell has increased the number of salespeople it employs, as fewer people are needed behind the scenes to handle outdated technology and data.

“One of the trends that we've seen across all of these is when you do AI right, you may or may not have more or less people, that's not really the goal. But what you do find is that the people that are in your organization shift to the core work: if you're in sales, you sell.”

There’s intense interest in replicating these results across Dell partner and customer organizations. Roese told ITPro that he spent much of 2025 helping customers to move AI projects into production and to validate what had been, up to this point, theoretical benefits.

“We track how many one on one customer conversations I have and it was hundreds, hundreds of meetings with chief AI officers, CIOs, boards of directors, CEOs about this topic so there's an insatiable appetite.”

FOLLOW US ON SOCIAL MEDIA

Make sure to follow ITPro on Google News to keep tabs on all our latest news, analysis, and reviews.

You can also follow ITPro on LinkedIn, X, Facebook, and BlueSky.