Expertise shapes how far workers trust and delegate to AI

A controlled think-aloud study examined how expertise and trust shape the way workers pass work to an AI assistant. Novices accepted AI output with little scrutiny. Experts checked AI output against their own knowledge and caught more errors before those errors carried forward. The gap matters for product design. A single review screen treats a novice and an expert as if they need the same thing. They do not. The novice needs more signal about where the AI is confident and where it is guessing. The expert needs enough transparency to apply their own judgment fast, without having to reconstruct what the AI did from scratch. The study also found that stated trust and actual behavior diverged. Workers who said they trusted the AI still hesitated before passing over consequential tasks. That hesitation is the moment where a person decides whether the AI's output is good enough to act on. Interfaces that rush past that moment, or that make review feel like friction, leave the worker without the footing they need to catch mistakes.