The vendor chose “agentic” as its word of 2025. For 2026, it has chosen “governance”.
“In the AI world, governance is becoming increasingly important because as you get through the learning curve of understanding the technology, and you want to put things into production, or you want to figure out how to get the best value out of AI, or how to reduce the risk, you need rules,” said Roese.
“Where we are today is a fairly bad state for external governance. There are over 1,000 governmental jurisdictions in the world that have created independent AI regulations, and none of them coordinated with each other,” he said.
He added that Dell was sounding a call to action for “governments around the world and the industries within each country to really spend more time and energy trying to get to more rational, harmonised governance from a regulatory perspective”.
Roese said industry and governments would not solve the issue next year, but it was important not to make things worse.
At the same time, internal governance around AI continues to be an area where many customers have yet to set out clear guidelines, but that will change going forward.
In the AI world, governance is becoming increasingly important, because as you get through the learning curve of understanding the technology, and you want to put things into production, or you want to figure out how to get the best value out of AI, or how to reduce the risk, you need rules
John Roese, Dell
“I’ve talked about this for a year and a half [since] we at Dell figured out early that having a top-down, highly governed approach to AI is the right answer, and it allows us to get into production with targeted, strategic projects and keep the rest of the noise away from distracting us,” he said.
“We are still finding that many enterprises have not achieved a strong level of governance. They do not know how to create prioritisation. They do not know how to say no to things. They do not know how to associate ROI [return on investment] with a use case so that we can prioritise it,” added Roese.
“We believe that as the technology continues to evolve, as new capabilities come out, governance inside of the enterprise will become a priority,” he said. “Enterprises will start to mature and realise that strong internal governance, a framework in which they execute their AI strategy, is essential for them to move forward, and we will see more and more of that activity occurring.”
Connected to that, Dell is also predicting the emergence of a knowledge layer sitting between systems of record – the customer relationship management (CRM) and enterprise resource planning (ERP) options – and AI compute.
“The part that is changing is in between the systems of record and the AI workloads that run on compute – there is a new layer of data management that we sometimes call a ‘knowledge layer’. The knowledge layer is the explicit place where vector databases, graph databases, knowledge graphs – all of the tools that organise data so that AI systems can use it – exist,” he said.
“Unfortunately, early on – meaning this year and even last year – was very ad hoc. If you needed a vector database, you just kind of attached it to your existing system of record. If you needed a graph database, you would go to wherever your databases were and run it,” he added.
Dell is expecting the knowledge layer to more closely align with the AI compute level, which is likely to be welcome news for those channel players that excel in that area.
“Our prediction is that when you think about the knowledge layer, it is an independent layer, and you have to make a decision – where does it belong? Does it belong with your systems of record, where the primary data exists, or should it be highly connected to the AI compute environment, the layer that actually uses the data? Our view is that having the knowledge layer as a separate layer that is closer to the AI compute side is more valuable than trying to embed all of these data structures into your traditional, non-real-time systems of record,” said Roese.
“That is a very significant change, and it will require people to rethink their data architecture and become much more explicit about the layer of technology necessary to serve data into the AI compute world as they move into more advanced chatbots, agentic systems and other technologies,” he added.
Roese made other predictions around AI agents, revealing expectations that the technology would unlock positive changes across organisations that had not been anticipated as the technology extends its presence in the workplace. He said next year should also include more of a focus on resiliency as customers look to ensure systems are robust.
