Unless those who have access to the agents discovering the advance vulns are few, and those who have access to the agents writing vulnerable code are many, right?

That's the same situation as before AI: many developers don't have time/resources/expertise to prevent the advanced exploits, and few very advanced players do.