In September, economists from Anthropic published a note titled Scenarios for our Economic Future, which outlines three potential scenarios for how artificial intelligence might affect the global economy by 2030. In the most extreme scenario, AI becomes more productive than humans on most intellectual tasks. This would lead to a 32.4 percent increase in gross domestic product compared to a world without AI, with annual economic growth reaching 15 percent and the economy doubling in size every four and a half years. However, this scenario also predicts a decline in salaries for knowledge workers by over 10 percent, with unemployment reaching 11.9 percent overall and 17.9 percent specifically for knowledge-based jobs. In this scenario, capital earns more than labor, with 54.8 percent of national income going to capital versus 45.2 percent to labor. The report highlights that while society would be wealthier than ever, fewer people would have jobs in knowledge-based fields, and unemployment would surpass typical levels seen during economic downturns. The authors of the note acknowledge that their model focuses on certain key economic forces while leaving out many other factors. Nonetheless, the report reflects the perspective of a company that is actively involved in AI development. According to the note, the share of labor in national income would decrease across the three scenarios, from 59.4 percent to 45.2 percent. This is not the view of a critic of AI but rather one of its creators. The note does not explain how AI could surpass humans in most intellectual tasks, but one possible explanation is that AI systems are trained using the knowledge and methods of human workers. When managers input their expertise into AI systems and refine the outputs, they essentially teach the machine how to perform those tasks. Similarly, when companies share their procedures and data with AI models, they are giving the systems the very tools they sought to enhance. This process increases AI's productivity but at a cost: the know-how that once defined the value of a job and the company is transferred to a system that everyone can access, including competitors. The issue of control over AI models is crucial. Some may argue that suppliers of AI systems commit to not using customer data to train their models, but this is not a guarantee. In 1998, two Stanford researchers, Sergey Brin and Larry Page, published a paper on search engines and warned that search engines funded by advertising would be biased in favor of advertisers and against users' needs. Despite this warning, they later created Google, a search engine funded by advertising. Similarly, in 2014, Facebook claimed it couldn't link WhatsApp users to Facebook accounts, but in 2016, it changed its terms to allow such linking. The European Commission later fined Facebook for misleading the EU about its capabilities, highlighting how commitments can change when incentives or power dynamics shift. This shows that written agreements may not be enough to ensure long-term reliability when one party can change its stance once the other becomes dependent. To address these concerns, companies should distinguish between the aspects of a job that define it and those that merely occupy time. The former should be handled by models that the company owns, rather than those shared with others. Possessing a model means ensuring that the data stays within a defined legal jurisdiction, that processing occurs within that jurisdiction, and that the model, its updates, and its evolution remain under the company's control. This is referred to as "triple sovereignty" and should be a key consideration when choosing AI suppliers. A model that a company owns can be tailored to its specific needs and can learn from its own data rather than from a shared pool. Over time, this gives the company a unique advantage, as its AI system becomes distinct from those of its competitors. Ultimately, the question raised by Anthropic's note is not whether AI will replace managers, but rather, to whom these managers will have taught their jobs in the process.