Do workers want the same things from AI across every task — or does it depend on how pointless the task feels?

This explores whether workers' general wishes for workplace AI stay the same across all their work, or bend depending on what a given task is like. No study in the corpus measures both at once. Put several studies side by side, though, and the answer looks like yes. The general priorities are real, but how much they matter seems to rise and fall with the task.

Start with the general view. When business users are asked what they want from AI agents, they name human control, reliability, awareness of context, and safety as practical requirements What UX principles do workplace users want in AI agents?. These sound universal. A study of workers rating 171 of their own tasks suggests they aren't. The more pointless or disconnected from company goals a task felt, the more willing workers were to hand it to AI, and the less human oversight they thought it needed Does perceived task meaninglessness drive worker desire for AI delegation?. So 'I want to stay in control' isn't a fixed setting. It loosens for work people don't care about. The pattern held across 22 occupations, which suggests the task matters more than the job title.

Real-world data on where workers have actually delegated work to AI points the same way. That delegation clusters in information-heavy work, and it follows what the technology can do rather than how widely people chat with LLMs. It also doesn't match older predictions that 'routine' jobs would be automated first Where have workers actually delegated tasks to AI?. Task traits that look like they should predict delegation, such as being repetitive, turn out to be weaker predictors than whether the AI can handle the information involved and whether the worker finds the task meaningful. On the capability side, agents complete only about 30% of tasks in a simulated workplace. They fail most often on tasks that need social interaction or specialist knowledge Why do AI agents fail at workplace social interaction?. That makes it reasonable to want tighter control over those kinds of tasks specifically.

There is a twist. Individual preferences, whether tied to a task or not, may matter less than people assume. In a study of engineers, company rules set how much control people kept over agentic AI before personal preference came into play. Those rules included mandated tools, approved-tool lists and data policies Does personal preference shape how engineers use AI tools?. Social pressure adds another filter. People expect to be judged less competent for using AI, so they hide it Do people fear judgment when they use AI at work?. A worker might be happy to delegate a meaningless task but still avoid doing it openly.

The takeaway you might not expect: asking people 'what do you want from workplace AI?' probably gives you an average over very different tasks. The more useful question is 'which of your tasks would you hand off, and with how much checking?' The answer depends partly on whether the work feels meaningful, which has little to do with how hard it is.