Developers evaluating AI code mostly just run it.

Two papers from the same week say this independently. A 100-person observational study had participants cycle through five AI suggestions varying in security and functionality across four C linked-list tasks. And 527 free-text responses from researchers who write code found over half described running the generated code as the validation step, with automated tests and peer review rare. The confidence finding inverts with experience: less experienced programmers trusted the AI more than themselves, experienced programmers the reverse, and evaluation confidence correlated with trust rather than with the rigor of the strategy used.