I like the first 5 minutes from 22 minutes. Its a great example of expertise. It does sound a little bit like a "grindset" clip... BUT I think instead it ended up being really positive about wholistic knowledge and experience. He has an interesting POV about knowledge and curiosity needing to be balanced with execution. The curiosity needs the willingness to start working, but if you aren't curious, you do the wrong things. Insightful.
Video 2Do you know how rediculous it sounds to hear "we will soon find every vulnerability in every system with agentic frameworks"????? That is such a BS claim. Just flat out. Because vulnerabilities are not just about the function of the system, but also about the creativity of the attacker and the upkeepers. You cannot expect any such project to have 100% coverage. That is just a fact of life.
Video 3Brother its just insane to hear people talk about this kind of thing. They really make the wildest claims that they ground in the idea that they know what the best practices are for using LLM development. And in the end its always just "idk ask chat/claude/codex to do it". ITS INSANE. That cannot work longterm. For millions of reasons. But most of all because it will eventually dilute itself. Its a well known phenomena that models that feed off the internet steadily decline after 3 months as it consumes more AI generated content because it is diluting its own weights with guesswork. The more repositories and security content that is AI generated, the worse the models will get at understanding security. Its just a fact of life. The best practice is not just "human in the loop" but "human DRIVING the loop". Again, going back to the knowledge versus experience conversation in video 1, you can ask for an LLM to execute a recipe and make you something, but you are going to make the wrong thing if you dont have the knowledge. The AI certainly does not.