AI has accelerated coding, and Sonar rebuilds confidence with verification loops
Sonar proposes a code confidence formula of "intent × guardrails × verification × evidence," embedding automated verification into three loops: agentic coding, CI verification, and code maintenance. The Carnegie Mellon University research it cites shows that public repositories adopting Cursor saw a 3 to 5 times increase in new lines of code in the first month, but the gains faded after about two months, with static analysis warnings increasing by 30% and code complexity rising by 41%. Sonar argues that having the model that generated the code check its own work is like "a student grading their own homework," and that verification must happen synchronously while the agent writes code.
Body
The English body is being prepared. You can open the original source.