AI governance often starts in the wrong place. The real question is not only which tools people use, but what decisions AI is allowed to influence — and who remains accountable when the output is wrong.