Static code analysis is great but it misses a lot. Agents that read code make guesses about how code would execute, Ito makes a guess and then verifies it with runtime evidence.