Does AI help more when workers choose to use it themselves, or when a company forces everyone to use it?

This explores whether AI's effects change when people choose to use it versus when the choice is made for them. The collection has no head-to-head study of optional versus mandatory use. It does show what happens on each side of that line, and together those findings point to one variable that matters more than the label: whether the person can check what the AI produces.

When AI is optional, people seem to sort themselves sensibly. Interviews with junior developers found that their main reason for picking AI or manual work was neither deadlines nor task difficulty. It was whether they could verify the output Do junior developers choose AI based on their ability to verify results?. They avoided AI where they couldn't judge the result and used it where they already had expertise. This matches a broader argument that AI does best on tasks whose answers are easy to check Does task verifiability determine what AI systems will learn to solve?. Given a free choice, individuals seem to apply a rough version of that rule themselves. One reading: optional use can look effective partly because people steer the AI toward work they can quality-check.

Mandates take away that self-sorting. A study of junior and senior engineers found that company rules, such as required tools, allow-lists and data policies, set how much control people keep over agentic AI before personal preference comes into it Does personal preference shape how engineers use AI tools?. Under those constraints, the novices swung between relying on the AI too much and avoiding it. They lacked the verification ability that the voluntary users above relied on, and the policy no longer let them route around the gap. A related cost appears when AI help arrives without being asked for: even correct suggestions can break a person's concentration and make their reasoning worse Does AI assistance always help reasoning or does it carry hidden costs?. Help you invited and help that interrupts you are not the same help.

Bans give the mirror-image result. In one master's program, a majority of 2025 applicants submitted essays that were likely mostly AI-written, despite an explicit prohibition. The AI improved essay quality, yet those applicants were admitted at lower rates than similar applicants who didn't use it Does AI essay use hurt admissions chances despite quality gains?. A rule doesn't decide whether people use AI. It decides how that use is judged. Readers and writers also disagree on when AI use should be disclosed: readers consistently consider it more necessary than writers do Do readers and writers differ on AI disclosure necessity?. So the people using AI and the people evaluating the work may not agree on what counts as acceptable.

The unexpected takeaway: the optional-versus-mandatory question may really be about verification and evaluation. Voluntary users protect themselves by staying inside what they can check. Mandates can push people past that line, and bans mostly change who gets penalized. If you want to explore further, compare the self-selecting developers with the engineers working under policy, which are the two closest studies the collection has to a direct comparison.