The AI Revolution

A reasoning model does not simply give the first answer that comes to mind. It works through a problem in intermediate steps, tries different routes and checks its own work along the way. That lets AI take on problems that were long out of reach: in 2025, AI systems reached gold-medal level on the problems of the International Mathematical Olympiad, the competition for the world's best young mathematicians. In a conversation, a model like this thinks with you: it puts forward ideas of its own, points out weak spots in an argument and raises counter-arguments. In organisations, it runs an investment through different scenarios, tracks down a bug in thousands of lines of code or sets a new contract against every earlier agreement.

Multimodal AI processes several kinds of information at once: text, images, sound and video. That means an assistant can answer questions about what someone shows it or lets it hear, without everything having to be typed out first. AI is no longer tied to the keyboard. A technician points a phone at an installation and hears what might be wrong. You talk to an assistant like this as you would to a person: you interrupt it, point at something and it responds straight away. During an inspection, AI compares camera footage with earlier recordings and flags changes that deserve a closer look. Personal assistants move with you from laptop to phone and remember what you were working on. AI becomes a colleague that watches and listens alongside you, wherever you work.

AI gives robots, cars and devices a brain. A robot, a self-driving car, a camera or a sensor then understands what is happening around it and can respond. That is called physical AI. A conventional industrial robot follows a fixed program; a robot with AI learns from examples: from videos, from people demonstrating a task and from practice in a simulated environment. Self-driving taxis already carry passengers without a driver in several cities, humanoid robots work on assembly lines in trials, and cameras with AI spot when a machine or a set of railway points needs attention. Speed, cost and working safely alongside people still limit their use; that is exactly where things are moving quickly.

A general language model is AI trained on vast amounts of text, and it knows something about many subjects. A specialist model is also tuned to one field: legislation and judgments, medical guidelines, technical manuals or an organisation's own documents. With access to those sources, it finds the relevant passages and shows where an answer comes from. Specialist AI goes beyond language. There are models that predict molecular structures, calculate next week's weather or spot abnormalities on a scan. In organisations, a model like this works alongside the professional: AI does the searching, reading and calculating; the professional weighs the evidence and decides.

Generative AI creates new content from an instruction in everyday language. Where that once meant text and single images, the newest models produce video with matching sound, voices in many languages and images that cannot be told apart from a recording. A scene now comes together without actors, voice actors, cameras, a location or a film crew. A training department makes instructional videos with spoken explanations in several languages. A designer shows a hundred versions of a package before a single one is printed. The next step: films and animations on demand, made to measure for the viewer. And world models: AI that simulates an interactive environment in which robots can also practise.

An AI assistant helps with a question or task; after that, the user chooses the next step. An AI agent is given a goal and takes the steps to reach it: finding information, opening a system, filling in a form, sending a message and checking whether it worked. An agent prepares an application from start to finish, instead of writing a single piece of text. Agents are quickly getting better at real computer tasks and are taking on ever longer sequences of work. What that means for the work, for oversight and for what an agent may do on its own is the subject of the Agentic AI keynote.

In many organisations, AI is still added to work that already existed: an assistant for email, a summary after a meeting. An AI-first organisation starts from the other end. AI and new technology do the work people need not do, including the repetitive and predictive work, and people focus on the work where they make the difference: judging, choosing and working with people. At a housing association, a repair request then moves from the first report to a booked technician without anyone stepping in, and the housing officer has time left for the resident who is stuck. Which tasks can AI take on, and which belong with people? Organisations are asking that question now, process by process. In McKinsey research, redesigning workflows in this way is strongly linked to the financial results organisations attribute to AI.

An AI tool can be available quickly; the work around it changes more slowly, and that is often down to behaviour. Habit, the fear of becoming redundant, the sense that expertise is losing value, no time to practise and doubt about what is allowed: they come up in many organisations. According to KPMG, a large share of employees already use AI, though by no means always openly. Where that use becomes visible, things often move faster. A colleague who shows how AI saved them an afternoon makes the possibilities concrete for the whole team. That is how AI grows from the ground up, through early adopters, from a one-off trial to a habit.

When AI models scenarios, predicts deviations in processes and shows decision-makers new connections, people's work shifts. They carry out less themselves and direct more: they set the task, assess the result and step in where the system lacks context. New roles emerge, such as the colleague who checks AI's work or the specialist who designs workflows with AI. People and AI form one team: AI prepares, and people bring judgement, human contact and accountability. Entry-level work changes too. Tasks such as finding sources and organising case files used to be how newcomers built up expertise, and those are exactly the tasks AI increasingly takes over. That is why new learning routes are emerging for people at the start of their careers.

Employees read the direction of an organisation from how its leaders behave, often more than from the strategy on paper. Leaders who set the example take the initiative: they use AI themselves, show what it delivers and make room to learn and try things out. The future then becomes something people build together. That room has limits: where mistakes have serious consequences, people practise first in a safe environment. The leader's own work changes too. AI brings more scenarios and signals into view; decisions about direction, people and values stay with those who lead the organisation.

AI literacy means knowing what AI can and cannot do, what can go wrong and how to use it safely, including with the data you put into it. Recognising a convincing answer is different from being able to judge what it rests on. For an employee, that is about what a prediction means; for a leader, about the assumptions behind a modelled scenario. The EU AI Act requires organisations that use AI to take measures that support the AI literacy of the people who work with it. More and more organisations train their people in this themselves, in a future academy where that knowledge grows along with the profession.

AI is being developed by large model builders, specialist companies, start-ups and research groups all over the world, and governments are investing in computing power and shared data. An AI ecosystem is the network of start-ups, scientists, partners and governments with which an organisation shares knowledge, data and solutions. Hospitals train a model together while patient data stays in-house, so it learns from data held by many institutions at once. Transport operators share planning data to avoid empty journeys. The pace of development makes that network more and more important: what has already been tried elsewhere does not have to be reinvented.