I just presented a new paper at @BrookingsInst on the "Vanishing Advantage of Specialization (new paper)". It explains why this will be the age of the generalist. We all understand that the key determinant of the economics of airlines is the utilization of the aircraft. The fuller the plane, the lower the cost per seat. The same mechanism explains the value of specialization. Acquiring a piece of knowledge is very expensive in terms of our cognitive effort. A specialist is a person who uses this knowledge sufficiently often to spread that cost over many uses. This explains outsourcing of "non core" capabilities in firms. A firm handles frequent problems internally and buys expertise outside for problems it encounters too rarely to justify maintaining the knowledge in-house. For instance, Microsoft has a large legal department, but it hired a corporate law firm to advise it in its purchase of Activision Blizzard. It is not because Microsoft lacks lawyers that it relies on outside service providers, but because it does not use sufficiently often the specialized knowledge of mergers to justify an entire department. Artificial intelligence reduces this fixed cost of acquiring knowledge and therefore also reduces the advantage of specialization. Suppose a capability (such as merger law) costs $100,000 a year to maintain, a firm faces the problem 10 times a year while the specialist sees it 1000 times, and suppose that briefing an outsider costs $2,000 per case. Since the specialist's advantage due to utilization is $9,900 per case, the firm buys the service from the specialist. Now suppose AI reduces the annual cost of maintaining the capability to $10,000 for both. The utilization advantage of the specialist is reduced to $990, less than the cost of asking for help across the boundary, so the firm prefers to acquire the capability itself. The specialist still spreads the cost over more uses; the saving from doing so is now less likely to cover the cost of crossing the boundary because the fixed cost is lower. Recent evidence shows workers using AI to cross their occupational boundaries. Chin and Richmond (2026) classify more than 800,000 work-related ChatGPT messages from users whose occupations are known. Each message is assigned to the occupation whose historical tasks look most similar. Most of the usage is generic, including what one would expect: help in writing, editing and planning represents 61.5 percent of all messages. But 16.8 percent of all work messages, and 43.5 percent of the ones which are specific to one occupation, concern tasks usually undertaken by a different occupation. Among the ones that cross occupational boundaries are calculating financial data, troubleshooting software, drafting promotional messages, and writing to government agencies. Note that the fall in the cost of acquiring a given capability did not begin with generative AI. Google and databases had already lowered these costs. Generative AI greatly accelerates this process. Consider lawyers, accountants, consultants and software workers: of the 337,000 jobs created between 2022 and 2025 by the four professions, 335,000 were outside their main specialist industries-- law is still being done, but you need less of a specialist law firm to do it. Professional-service firms can be squeezed in the middle between model producers and their clients. The experts can be squeezed from above, by model makers or labs who supply the general knowledge at scale, and from below, by their clients who can apply this cheaper knowledge themselves. The specialist's comparative advantage lies in expertise that clients use too rarely to maintain internally and that remains costly to acquire even with AI, including the expertise required to check and verify a solution. Thanks to Jan Ebberly and @JonSteinsson for inviting me to do this: Summary and links to paper and slides here: brookings.edu/articles/the-v… Paper: brookings.edu/wp-content/upl…