Daniel Kokotajlo, a former OpenAI researcher who worked in the governance division of OpenAI, has become a prominent voice in discussions about the future development of artificial intelligence. In a recent interview, he described scenarios that he believes could emerge as AI systems become increasingly capable. Kokotajlo has said that when he left OpenAI, he was prepared to forgo equity worth approximately $2 million rather than accept restrictive non-disparagement terms. OpenAI subsequently said it would no longer enforce the relevant provisions and would remove them from its exit paperwork. His forecasts remain his own assessments rather than established outcomes, but they raise an important question: what could happen if, within the coming years, AI systems become capable of performing a much larger share of the intellectual tasks currently carried out by people?
Kokotajlo's concerns are not limited to an extreme scenario in which advanced AI systems become difficult to control. Another part of his argument focuses on what may happen before such a point is reached: AI systems gradually taking on a growing share of the tasks that support the modern economy. According to the scenario he describes, the transition would not necessarily begin with robots replacing workers overnight. It could instead take place gradually through the software used in everyday work. A programmer may use an AI agent to write, test and review code. A lawyer may use a system capable of processing large volumes of documents. Medical professionals may use tools to analyse information and research, while managers may use AI to summarise meetings, prepare forecasts and organise information. Humans may initially remain closely involved in these processes, while the degree of automation could increase as the technology develops.
This raises broader economic and social questions. For centuries, employment has provided not only income but also, for many people, a source of identity, status and social participation. If AI systems eventually allow organisations to perform some activities with fewer human inputs, the implications could extend beyond productivity. Policymakers, businesses and society may need to consider how income, employment and working time evolve in response. Could shorter working weeks become more common? Could proposals such as universal basic income receive greater attention? Could people spend more time on education, family or creative activities? At the same time, there are questions about whether the economic value created by advanced AI could become concentrated among a relatively small number of companies that control important models, computing infrastructure or data. These remain open questions rather than predetermined outcomes.
This is also one reason competition among companies such as OpenAI, Anthropic, SpaceXAI, Google and Meta extends beyond the development of consumer chatbots. In Kokotajlo's scenario, increasingly capable AI systems could assist with scientific research, software development and the development of future AI models. If one organisation were able to deploy systems capable of performing substantial amounts of research or engineering work, Kokotajlo argues that this could potentially accelerate its development cycle relative to competitors. Such a system might support faster experimentation, product development or model improvement. Whether this type of self-reinforcing technological advantage will develop, and how significant it could become, remains uncertain.
From a capital-markets perspective, the development of AI highlights the different layers of companies involved across the technology's value chain. NVIDIA (NVDA) supplies processors widely used for AI training and inference, while Taiwan Semiconductor Manufacturing Company (TSM) manufactures many of the advanced semiconductors used by the industry. Broadcom (AVGO) and Marvell Technology (MRVL) participate in areas including custom semiconductor and networking technologies that support large-scale computing infrastructure. Oracle (ORCL), Amazon (AMZN), Microsoft (MSFT), Alphabet (GOOGL) and CoreWeave (CRWV) are among the companies providing cloud or data-centre infrastructure used for AI workloads. Other parts of the supply chain include semiconductor-equipment companies such as ASML, memory manufacturers such as Micron (MU) and SK hynix, and a wide range of infrastructure and component suppliers. If demand for AI computing capacity continues to increase, developments across these different parts of the supply chain may remain relevant to the industry's ability to expand. This does not mean that increased AI demand will necessarily translate into higher revenues, profits or share prices for any particular company.
The next layer includes companies seeking to integrate AI capabilities into products and business processes. Meta (META), for example, continues to incorporate AI functionality across its consumer and business platforms, while Palantir (PLTR) provides software used by governments and organisations for data analysis and operational decision-making. Other software companies are developing AI-related tools for areas such as customer service, data analysis, cybersecurity, sales and product development. If AI capabilities improve at a pace similar to the scenarios described by Kokotajlo, companies may increasingly examine how automation affects workflows, staffing requirements, decision-making processes and operating costs. However, the extent of these changes will depend on technological performance, regulation, implementation costs, customer adoption and many other factors.
Companies that benefit from the early stages of AI development may not necessarily retain the same position as the technology evolves. If AI reduces the cost of producing software, content or analysis, some existing software business models could face greater competition or pricing pressure. Functions that are currently sold as standalone products could potentially become features within broader AI systems. At the same time, businesses with proprietary data, established customer relationships, recognised brands, specialised expertise or physical infrastructure may have different competitive characteristics. Companies whose services rely heavily on repeatable intellectual tasks may also experience changing competitive conditions. The development of AI should therefore not be viewed only as a growth story; it may also change how value is distributed across industries and business models.
One of the central ideas in Kokotajlo's argument is that major changes may not arrive through one single dramatic event. Instead, they could emerge through a sequence of smaller technological improvements: a more capable AI assistant, followed by an agent able to complete larger tasks, and potentially systems capable of coordinating multiple specialised agents. Over time, such developments could influence the way people work, learn and interact with technology. Kokotajlo's forecasts may ultimately prove too aggressive, and technological progress could be slower or more constrained than he expects. Regulatory, economic, technical and social factors could all affect the pace of adoption. Nevertheless, even partial advances in the direction he describes could raise important questions about who develops and controls advanced AI systems, how the economic value they create is distributed, and how the role of human labour may evolve as increasingly capable forms of artificial intelligence become available.
This material is provided for educational and informational purposes only and does not constitute investment advice, investment research, a personal recommendation, or an invitation or solicitation to buy or sell any financial instrument. Past performance is not a reliable indicator of future results. Investing involves risk, including the possible loss of capital.
References to companies and securities are included solely to illustrate businesses participating in different areas of the AI ecosystem and should not be interpreted as investment recommendations. Forward-looking statements, forecasts and scenarios relating to artificial intelligence are inherently uncertain and may not materialise.
-(1)%201%20(4).png)