Matt Shumer is not a futurist in search of a compelling headline. He is the founder and CEO of OthersideAI, the company established in 2020 that developed HyperWrite, one of the first artificial intelligence-powered writing assistants. He builds products and invests in the sector. Shumer is therefore both a witness to the acceleration and an entrepreneur operating in the market that this acceleration promises to expand.

In February 2026, he published “Something Big Is Happening”, an essay centred on a precise prophecy. According to Shumer, the world today is in the same psychological condition it experienced in February 2020. At that time, a small number of people were watching the news from abroad with concern, while the majority continued to work, travel and make plans for the future. The alarm appeared excessive until, within three weeks, normality changed shape. Something similar, he argues, is now happening with AI, but on a larger scale and with permanent effects: not a crisis to be overcome, but a transformation of cognitive work.

The significance of his argument lies in the classification error that accompanies almost every major transition, when a new phenomenon continues to be placed within categories created for the one that preceded it. Artificial intelligence is still treated as a collection of tools to be added to the company; meanwhile, it is becoming the environment in which the company creates knowledge, organises work and decides what it considers true, urgent or advantageous.

Judging by the figures, adoption already appears to be complete; it is the depth of the change that remains far less certain. In 2025, 88 per cent of the organisations surveyed reported using AI in at least one function, while 70 per cent were employing generative systems. The technology had reached an adoption rate of 53 per cent in three years, faster than either the personal computer or the Internet. The use of autonomous agents, however, remained marginal across almost all corporate functions, while many initiatives had not yet moved beyond the pilot stage. These figures primarily describe ease of access. Between purchasing a subscription and redesigning an organisation lies the same distance that separates an installed machine from a factory that has genuinely transformed its processes, responsibilities and working criteria.

The speed at which system capabilities are increasing cannot be dismissed as commercial rhetoric. METR measurements indicate that, since 2019, the time horizon of software and research tasks that models can complete with a given level of reliability has approximately doubled every seven months. The finding concerns clearly defined and verifiable activities and does not justify the conclusion that every profession is approaching automation. It nevertheless remains significant for those responsible for managing a company, because it describes software that is now capable of linking actions together, using tools, identifying errors and pursuing an objective. The transition from recommendation to execution introduces into the organisation a problem that productivity alone cannot fully describe: the distribution of authority.

The scale reached by this transformation is evident in the financial sector. JPMorganChase has made its LLM Suite available to more than 200,000 employees within a controlled environment designed to protect corporate data and client information, and has identified more than 500 artificial intelligence use cases already in production. The number of texts generated or minutes saved explains only part of the phenomenon. An internal platform of this size places a common cognitive layer between people, data and processes; it changes the way information is searched for, reorganised and delivered to those who must work or make decisions. In such a structure, AI governance cannot be a policy added at the end of the process. It coincides with the organisational architecture, the management of access rights and the definition of what a machine is permitted to know and do.

In manufacturing, the issue takes a less visible form because it becomes part of the ordinary continuity of industrial operations. Siemens has brought its Industrial Copilot from the engineering office to the factory floor, making it available to a potential audience of more than 120,000 engineers and involving over one hundred customers across Europe and the United States. The system generates and corrects automation code, supports maintenance, queries industrial data and shortens the path between a physical problem and the knowledge required to solve it. The advantage arises from the interaction between the model, the industrial plant and the technical knowledge accumulated by the company. Greater efficiency ultimately also affects the development of experience: when a solution is routinely proposed by the copilot, it becomes less clear through which continuing processes people develop the ability to recognise an incorrect one.

The International Labour Organization estimates that approximately one worker in four is employed in an occupation exposed to generative AI to some degree, while specifying that the most likely outcome is the transformation of tasks rather than the uniform elimination of jobs. For companies, the significance of this finding is not limited to the number of employees they will require. It also concerns the way professionals will continue to be trained. Many senior skills originate in years spent performing simple, repetitive and sometimes imperfect tasks, during which people learn about exceptions, contexts and consequences that no job description can record in their entirety. Automating that stage may improve financial results in the short term while weakening, without immediately visible effects, the process through which the organisation produces judgement.

Artificial intelligence depends on advanced semiconductors, data centres, energy, cloud services, data and foundation models. This is not a supply chain distributed evenly across the world. Some stages are difficult to replace, a small number of operators control decisive portions of the chain, and the most advanced capabilities remain concentrated in a limited number of territories. Stanford’s AI Index 2026 observes that frontier models are produced primarily in the United States and China; during the same period, a growing number of governments are investing in national computing capacity. For a European company, the choice of a platform therefore entails dependence on the jurisdictions, infrastructures and industrial policies that will ultimately hold part of its operational memory.

The European response is proceeding on two levels that do not always advance at the same speed. From 2 August 2026, the European Commission and national authorities began applying new provisions of the AI Act, including transparency rules for certain interactions and for some artificially generated content. At the same time, the InvestAI initiative aims to mobilise €200 billion, including a €20 billion fund dedicated to artificial intelligence gigafactories. Regulation can protect the market and establish conditions of use even for systems developed elsewhere. A country’s position within the global hierarchy of knowledge, however, will depend on the actual availability of computing capacity, expertise and models. Sovereignty does not necessarily require autarky, but it does require the existence of viable alternatives.

For boards of directors, the priority should not be to adopt AI as rapidly as possible. Determining which decisions may be delegated also means identifying which decisions must remain verifiable and who must be held accountable for their consequences. System inventories, records of automated decisions, thresholds for human intervention and assessments of dependence on suppliers have now become part of ordinary risk management. Metrics also require a broader scope: alongside cost savings, they must take into account errors, the loss of skills, the concentration of information and potential reputational consequences. An algorithm can optimise whatever it is given; the decision about what deserves to be preserved remains part of the company’s strategy, even when execution takes place in milliseconds.

The transformation predicted by Shumer is unlikely to take the form of a single recognisable event. It is more likely to proceed through individual delegations that appear reasonable when considered separately: greater speed, greater consistency, lower costs and improved access to information. Taken together, however, they may alter the nature of the organisation and its relationship with its own knowledge. The company of the next decade will not be defined solely by the amount of artificial intelligence it purchases, but by its ability to incorporate it without losing memory, responsibility and autonomy of judgement. Otherwise, the risk is that an increasing share of corporate intelligence will be entrusted to infrastructures that the company uses without truly governing them.