Improving AI agents through better evaluations - InfoWorld