Infornomics

Since the version I last shared, I made four changes. - First, I incorporated the publicly available usage data from Anthropic and OpenAI to extend our current "capabilities only" bottleneck pressure score to also account for usage, and ranked occupations by this usage-adjusted pressure index. The central task-level result holds with both usage sources: the same task faces different pressure across jobs. The task rankings remain similar, although the occupation rankings change substantially. For example, writers and computer programmers rise to the top under the usage-adjusted construction with Claude data. While much of their work is rated AI-feasible, accounting for usage leaves more of their work outside AI use, moving them closer to the even split where occupation-level pressure peaks. - Second, I added an empirical prediction. I show that bottleneck pressure also predicts faster posting growth after 2022, beyond what linear exposure or observed AI use would predict. - Third, the same AI-infeasible task faces more pressure, on average, in senior postings than junior ones within the same occupation in every year since 2018, and about 6.5 percent more in the first half of 2026. This reflects how jobs are designed along the career ladder. Whereas senior postings surround the same hands-on or interpersonal tasks with planning and scheduling work that AI CAN speed up, junior postings carry much more hands-on physical work that remains AI-infeasible. The gap reverses in information-intensive jobs such as accounting and credit analysis, where junior work is already mostly AI-feasible. - Fourth, mean occupation bottleneck pressure rose 6 percent from 2018 to the first half of 2026, holding AI feasibility fixed, with nearly all of the rise by 2023. About 75 percent of this rise reflects advertised task mentions shifting away from low-pressure occupations rather than pressure rising within them. Basically, advertised work is shifting toward occupations that mix work AI can do with work it cannot (yet). In the abstract below, the first two results are highlighted in purple and the third and fourth in teal.