Kathy Gibson reports – Agentic artificial intelligence (AI) became the big buzzword in 2025, but organisations will only start to realise real benefit from the technology in the year ahead.
That’s the word from John Roese, global chief technology officer and chief AI officer at Dell Technologies, who cautions that a lot of what we call agentic AI is not yet scratching the surface of what’s possible.
He believes there are some critical foundations that need to be laid before the world of agentic AI becomes a reality, starting with the often-neglected area of governance.
Both external and internal governance are going to be critical, and Roese urges governments and organisations to pay attention.
“As you want to put things into production, you need rules, both external and internal,” he explains.
“But where we are today is quite a bad state in terms of external governance. There is no co-ordination and, as a practitioner, it can be hard to negotiate.
“This is a call to action for governments and industries to aim for more harmonised governance.
“We won’t solve this in 2026, but we can make sure it isn’t made worse.”
We must focus more on real governance, Roese says. Without it, we will end up with uncertainty that will slow the adoption of practical and valuable enterprise AI.
“Our specific ask of both the public and private sector is to develop governance for the enterprise market in collaboration with the actual enterprise market ecosystem (real enterprises and real enterprise technology suppliers).”
At the same time data management – the true backbone of AI innovation – has to be updated.
In 2025, we have become comfortable with AI compute, Roese explains – but this is just one layer in the AI picture.
One of the other layers is the system of record. These won’t change in the AI era, but are a valuable source of information.
“What is changing is a new layer of data management we call knowledge layer. This is where the vector databases, graph databases and knowledge graphs exist.
“This has been pretty ad hoc up to now, but organisations are going to have to decide where this belongs. This is a significant change and will require people to rethink their data architecture.
“We expect the idea of a separate knowledge layer will happen in 2026.”
Roese’s third prediction revolves around agentic AI as the new operations continuity manager.
He explains that true autonomous agents are different from chatbots; they are a software system with multiple components.
“An agent is also not a large language model (LLM). It uses an LLM, but also uses specialised knowledge in the form of knowledge graphs. This is where we store long and short term memory.
“In addition, autonomous agents are able to perceive the world around them and act on it. They are able to read information from other data sets; and to use tools – both physical or IT tools.
And they are able to talk to each other and collaborate, either internally or with other organisations. This is a new capability that is real now and will become more significant.”
Given these capabilities, Roese warns that users of these agents – no matter what you think they will do – will underestimate what is coming.
“As we bring agents into our systems we find that they not only provide set of tools, but change the way the work is done,” he says.
“Sometimes they started out as a tool, but ended up as the continuity manager.”
Another way that agents can surprise is in scaling the capabilities of an expert, so it is available much more widely in the organisation.
“At the other end of the spectrum. autonomous agents can reduce cost and complexity that would be required if people were employed to do the same job.”
Roese’s prediction is that agents will change the way organisations operate, and help people to be better.
“We think there will be a lot of surprises,” he adds.
His fourth prediction is that AI factories will redefine resiliency and disaster recovery
“As AI becomes embedded in core business functions, continuity becomes non-negotiable,” he says. “So AI infrastructure will evolve to prioritise resiliency, redefining what disaster recovery means in an AI-driven world. T
“he focus shifts from simply backing up systems to ensuring AI capabilities remain operational, even if primary systems go offline.”
Today we don’t have a great record of building cyber resiliency into AI factories. Roese adds, but 2026 should see more focus going into data resiliency.
“We could see the emergence of logical AI factories; or we could see autonomous agents handling failover and resiliency. Or traditional tools may evolve to protect knowledge graphs and vector databases.
“The prediction is that this will be a very active space, although we don’t know what the end looks like.”
Roese’s final prediction for 2026 is that sovereign AI will accelerate the building of national enterprise infrastructure.
He says governments are going to become more involved in providing or collaborating in data centres.
“As AI becomes critical to national interests, we’re seeing the rapid rise of sovereign AI ecosystems. Nations are no longer just consumers of AI technology; they are actively building their own frameworks to drive local innovation and maintain digital autonomy.”
This shift is reshaping AI infrastructure planning, he adds, with AI compute, data storage and management playing pivotal roles in safeguarding and localising sensitive information.
“Enterprises will increasingly adapt to these sovereign frameworks, scaling their operations within regional boundaries. This creates a powerful new wave of localised innovation that has a real-world impact on citizens and economies.
“By keeping data within national borders, governments can shape their public services and enterprises can tap into new indigenous infrastructure while aligning business objectives to national industrial policies.
“This is a foundational change that moves AI from a global concept to a powerful local reality.”
Roese believes there will be a lot of activity in the quantum space during the coming year, although we probably won’t see any major breakthroughs.
“Quantum continues to expand and improve,” he says. “We are closing the gap to being able to use it in practical ways.
“There have been huge advances. I don’t believe we will see gigantic quantum breakthrough, but we are closer to convergence than most people think.
“And now AI is helping to build algorithms and build new systems.”
]