“Thomas Dohmke predicts that even in three to four years, AI agents will still be unable to fully handle the complexity of large-scale software projects, requiring human oversight.”