Why does AI hit desk jobs — reading, writing, deciding — harder than hands-on, routine work?

This explores why AI lands hardest on information-heavy jobs, and what that exposure actually looks like once it arrives. The most direct answer in the collection is that exposure follows what the technology can do, not how routine the work is. When researchers tracked where workers have actually handed tasks to AI inside structured workflows, the tasks clustered in information-intensive occupations and lined up with AI's technical capabilities. They did not line up with older predictions that automation hits repetitive, routine tasks first, and they did not simply follow how widely people chat with LLMs Where have workers actually delegated tasks to AI?. The study also found that the wage pattern flips at the advanced-degree level, so highly educated workers are not shielded the way earlier automation waves might suggest. Put simply, AI is good at manipulating text, data and documents, so the jobs made of those materials are the ones it touches.

The corpus goes further and asks which part of an information job is exposed. Narayanan and Kapoor split knowledge work into three layers: deciding what to do, executing it, and delivering it. They argue AI compresses mainly the middle layer, while the decide and deliver layers hold steady or grow Does AI really compress all layers of knowledge work equally?. That helps explain a puzzle. Translation and legal work are heavily exposed, yet employment there has stayed stable or expanded. A related labor study points to the same mechanism. When exposure is concentrated in a few tasks rather than spread across a whole job, workers can move toward the tasks AI doesn't touch, and the net effect on employment stays modest Does concentrated AI exposure enable workers to adapt and reallocate?. So how exposure is distributed across a job's tasks may matter more than how much exposure there is overall.

The less obvious finding is that 'information work' doesn't spread its exposure evenly across the people who do it. In male-dominated occupations, AI exposure concentrates among high-skilled, high-paid workers. In female-dominated occupations, it spreads across every skill and wage level, so lower-paid women end up just as exposed with fewer resources to adapt Does AI exposure hit low-wage workers harder in some fields?. Exposure also depends on the employer. More AI-exposed firms replace freelance labor with AI tools faster and more cheaply than less-exposed firms, which suggests that building AI capability inside a firm pays off with scale Do firms substitute labor for AI at different rates?. Acemoglu, Autor and Johnson argue that firms lean toward automating expertise rather than creating new tasks for workers because automation pays better, so the direction of exposure is partly a business choice and not only a technical fact Why do firms build automating AI instead of pro-worker AI?.

Finally, exposure is a lived experience as well as an economic measure. Gallup's panel of 30,000 U.S. workers found that daily AI users report more than twice the fear of job elimination that infrequent users report, though supportive managers shrink that gap noticeably Does frequent AI use make workers fear job loss more?. Many information workers also hide their use. In one interview study, most said AI saved them time, yet about 70% concealed or downplayed it Why do workers hide productivity gains from AI use?, and experiments show users expect to be judged less competent if they disclose it Do people fear judgment when they use AI at work?. Even 'safe' augmentation carries risk: leaning heavily on AI agents can slowly wear down the skills workers need to supervise them Does AI augmentation protect workers from skill erosion?. One caveat: the collection shows clearly that exposure follows capability, but it says little about which specific features of information tasks make them easy for AI to take on. That deeper 'why' is still mostly open here.