> AI agents can now autonomously write code, run tests, fix failures, and iterate, increasingly for hours without human input.

Citation needed. How correct is the code generated by AI agents ? Can an AI agent test its code ? Did anyone proved this ?