AI isn't automating the repetitive jobs we expected — it's taking over knowledge work instead, and even advanced degrees aren't safe.

This explores where workers are actually handing tasks off to AI (built into real workflows, not just asked of a chatbot) and why delegation lands in some kinds of work and not others. The clearest evidence is that delegated AI work concentrates in information-intensive jobs, and it follows what the technology can actually do more than how popular chat-style LLMs have become Where have workers actually delegated tasks to AI?. This upends an older expectation. Earlier automation forecasts pointed at routine, repetitive tasks. The real pattern points at knowledge work, and the wage relationship flips at the advanced-degree level, so the most highly educated workers are not simply the most protected.

Why information work? The corpus has an answer from an unexpected place: research on how AI agents should hand work to each other. That work names eleven features that decide whether a task can be delegated at all. The most basic one is verifiability, meaning whether anyone can check that the output is right What makes delegation work beyond just splitting tasks?. Information tasks like drafting, summarizing, coding and analysis tend to produce outputs a person can read and check. A second clue comes from a study of people using an AI agent. Trust dropped sharply for tasks that were irreversible and visible to others, like sending an email. It did not drop for tasks that were high-stakes but fixable What makes people distrust AI agents they delegate to?. Taken together, these suggest delegation spreads where mistakes can be caught and undone. A lot of information work fits that description.

Concentration doesn't automatically mean jobs disappear. A firm-level analysis from 2010 to 2023 separates how much AI exposure a job has from how spread out that exposure is. When AI touches a few of a worker's tasks, they can shift their time to the tasks it doesn't touch, and that largely cancels out the job losses. When exposure is spread thinly across everything a worker does, there's no part of the job left to move into Does concentrated AI exposure enable workers to adapt and reallocate?. So the useful question is not only which occupations are exposed, but how much untouched work each one still has.

How workers feel about this is a different story from what the job data shows. Anthropic's Economic Index found that the people who delegate the most to Claude are the most optimistic about their careers and say their skills are gaining value Does delegating work to AI actually damage worker skills?. That is a correlation within Anthropic's own users, though, not proof that delegation helps. A separate finding adds a caveat: AI productivity gains show up when people apply skills they already have. When workers use AI to learn something new, the gains disappear and learning suffers When does AI actually boost worker productivity?. One plausible reading is that delegation pays off for experienced information workers handing off work they could do and check themselves, and pays off less for people still building that expertise. Even then, the payoff may be less work rather than faster work. One study found AI-assisted interaction cut clicks and scrolling but didn't shorten the time it took to finish a task Does chat delegation actually save time on task completion?.

A gap worth naming: the corpus shows where delegation concentrates and gives strong clues about why, but it doesn't offer a detailed breakdown by specific occupation. The link between checkable outputs and where delegation happens is a synthesis across these notes, not something any single study tests directly.