Security executives are concerned about flaws in AI agents but also eager to see them replace humans in some roles, according to a new report.