Google’s engineering benchmark finds top AI agents scoring just 50.6%

AI's current limitations in meeting stringent engineering constraints highlight the need for further development before unsupervised deployment.