Developers hand AI more work when trust is high, study finds

A mixed-methods study combined interviews with statistical analysis to ask how developers decide what to hand off to an AI coding assistant. Trust was the strongest predictor. Developers with higher trust handed over larger, more complex tasks. Developers with lower trust kept the AI on narrow, well-bounded work. Trust was not fixed. It grew when the AI explained its reasoning and shrank when the AI produced confident-sounding wrong answers. Developers described a cycle: give the AI a small task, inspect the result, adjust trust up or down, then decide what to give it next. That cycle is the real interaction, and most tools do not support it. For product teams, the implication is concrete. A developer who cannot see why the AI did something cannot judge how much to rely on it. They keep the AI on small tasks as a result. Tools that surface reasoning, flag uncertainty, and make results easy to inspect give developers what they need to widen the work they hand over. The study treats trust as a design variable, not a personality trait.