AI gets delegated real work mostly in information-heavy jobs — why there, and not where automation forecasts said it would?

This explores why the tasks workers actually commit to AI, building them into structured workflows rather than trying them out in a chat window, pile up in information-heavy jobs. The most direct answer in the corpus is short: delegation follows what the technology can do Where have workers actually delegated tasks to AI?. It tracks technical capability more closely than it tracks how many people use conversational LLMs. That finding cuts against two common intuitions. Older automation forecasts predicted that routine tasks would go first. Delegated AI doesn't follow that line. And the wage pattern reverses at advanced degree levels, so the most credentialed information workers aren't the ones you'd expect to be shielded.

Capability alone doesn't explain why information work has so much that can be handed off. One useful lens comes from Narayanan and Kapoor, who split knowledge work into three layers: deciding what to do, executing it, and delivering it Does AI really compress all layers of knowledge work equally?. AI mostly compresses the middle layer. Information jobs happen to have a large execution layer that can be pulled out and handed over: drafting, summarizing, translating, searching. Their examples, translation and legal work, show steady or growing employment despite large AI gains. So high delegated exposure in a job doesn't mean the job is shrinking.

A second lens is trust. In a small study of students using a general-purpose agent, people withdrew trust when actions were irreversible and visible to others, like sending an email. High stakes alone didn't trigger it What makes people distrust AI agents they delegate to?. The corpus doesn't test this directly, but a lot of information work produces drafts and analyses a person can review before anything leaves their hands. That may be part of why it's the comfortable place to delegate first. It also fits the argument that risk grows with the autonomy you give an agent Does AI risk increase with the autonomy we give it?. Delegating bounded information tasks keeps a human at the decision point.

The part you might not have thought to ask about is that concentration matters as much as volume. A study of firms from 2010 to 2023 found that high average AI exposure reduces demand for labor. But when exposure hits only a few tasks within a job, workers move their effort to the tasks AI isn't doing, and net employment effects stay modest Does concentrated AI exposure enable workers to adapt and reallocate?. Anthropic's data adds a self-reported layer: its heaviest delegators are the most optimistic about their careers Does delegating work to AI actually damage worker skills?. That result is a correlation within Claude's own users, though, not evidence that skills actually hold up.

One caveat: the measured pattern may undercount some delegation. Across four experiments, people expected to be judged less competent for using AI and were less willing to tell managers about it Do people fear judgment when they use AI at work?. If that holds in workplaces, how visible delegation is may vary by job culture as well as by what the AI can do. The corpus has one primary source on the concentration pattern itself. The explanations above are the strongest adjacent evidence, not settled causes.