A major AI capability uplift is coming - are businesses ready?
Artificial Intelligence (AI) will soon be able to act on its own – observing a situation, deciding how to respond, and learning from the outcome, without waiting to be prompted. But businesses could miss out on the benefits, because of their internal processes, warns Milind, Principal Scientist at Mercedez-Benz Singapore.
Speaking ahead of The AI Summit, Milind says the rise of this ‘sovereign machine’ could generate millions of dollars in productivity gains. Like many commentators, however, he believes the tendency for companies to treat AI and business strategy as separate concerns will prevent it from being deployed to its full potential.
“We often see business leaders delegate AI strategy to IT teams, which makes sense from a technical perspective. But business strategy today increasingly hinges on technical capability – so we see the best results when business and AI strategy are designed together. For that to happen, we need AI fluency right across the C-suite,” he said.
“For example, one use case might call for an overhaul of staffing or recruitment strategy, which sits well beyond the remit of an IT team.”
As machines become more sovereign, this separation of business and IT strategy will matter more, given the extent to which new AI models can reshape day-to-day operations.
“The sovereign machine is always on. It can observe its environment, understand the situation, decide how to act – or when to hold back – then act, observe the outcomes, remember what happened, and continue the loop. Unlike today’s tools, it does not switch off after answering a single question.
“This is an enormous departure from today’s AI. Right now, mathematicians use AI to help prove or disprove theorems. It can take ideas from one field of mathematics and apply them to another, or combine them to find a solution.
“The more important shift will be when these systems are starting to work things out for themselves. They can reason towards a solution in situations where they have never seen relevant past data – and although that might sound abstract, it has real consequences for everyday work inside organisations.”
History repeating itself
Milind says that treating transformative technology as an add-on has failed companies before, drawing a parallel with the slow – and initially botched – arrival of electricity in manufacturing.
Thirty years after the first commercial power station was commissioned, around 95 percent of new manufacturing capacity was still steam-based – and many companies that did make the switch failed to see a return on investment.
“People couldn’t work out how to use electricity on the factory floor, and those who bought electric motors often treated them as a straight swap for a steam engine,” he said.
“It wasn’t until firms began reengineering the entire plant around electricity that the technology paid off. You could now put a small or large motor at each workstation, sized to the task and switched on or off independently. That meant redesigning the workflow from the ground up.”
The same logic applies to AI, he argues.
“It makes genuinely new things possible – things no person or earlier machine could do – which is precisely why the transition cannot be led by IT departments.”
The hesitation will make less sense
While attitudes towards AI are becoming more open, a large gap remains between what the technology can do and how far companies have deployed it – a gap Milind believes will make less sense after the arrival of the sovereign machine.
“Some of the hesitation we see today is warranted,” he said. “Large language models are stochastic by nature, meaning they can give inconsistent output. Ask the same question ten times and you might get ten slightly different answers, only eight of which are right. That doesn’t sit well in business, where you need reliability.
“But the newer models of the last six months are increasingly much more reliable, and that will keep improving. We have already moved from co-pilots to workflow orchestration and general-purpose agents that handle multi-step work dependably, and we are now moving towards autopilots – systems that are always on. In near future, these agents will have capabilities that look like science fiction today.”
Always-on agents
Always-on agents – such as Google’s Gemini Spark and Microsoft’s Scout – are advancing quickly, Milind said, and could soon change how organisations onboard AI.
“These are a genuine change of model. Normally, you prompt an AI: it wakes up, gives you an answer, and shuts down. If it is doing something wrong, it is on you to catch it. An always-on agent is different – it is watching continuously, it holds a set of high-level goals, and it weighs the implications of what it sees. If you are about to do something that works against those goals, it can flag or stop it without being asked.”
Crucially, newer systems may soon also have temporal awareness – understanding how a situation unfolds over time. This means that the technology could soon be trusted to operate in new contexts.
“They are slowly but surely becoming genuinely self-learning, and the way they handle memory is improving dramatically,” Milind said. “There are two sides to this. The first is working memory: holding the right context, retrieving it when needed, and judging what matters and what doesn’t. The second is durable learning: turning experience into knowledge the system keeps and builds on. Today’s agents mostly do the first. The ones coming next will do both – they won’t just recall, they will continuously learn.”
Getting ready
The technology available today is already highly capable, and the pace of improvement is increasing. In this landscape, Milind says now is the time for businesses to prepare.
“We have to start now, so that we are ready to onboard these systems and put them to work. They will bring real risks too, so we need to think carefully about the architecture that keeps those risks under control.”
Learn more about what businesses can do to prepare for this shift at the upcoming AI Summit, where Milind will discuss:
• Why AI transformation is a board-level responsibility, not an IT project
• What “sovereign” machines are – and the practical capabilities that set them apart
• How continuous learning and memory will change the work agents can be trusted to do
• The architecture and governance needed to deploy always-on agents safely
Learn more and register for your tickets here.
The views and opinions expressed in this article are Milind's own and do not necessarily reflect those of Mercedes-Benz