If agents write most software, they could enable enforcement of better security architecture over time and perhaps reduce vulnerabilities.

“The only tactical thing I, I think is an interesting idea is the idea that, um, you could over time If, if it is going to be mostly agent-written code, you could enforce better security architecture, um, and, you know, perhaps fewer vulnerabilities or a lot of people where, you know, formal verification has always been enormously niche, right?”