LLMs give infinite mana, not infinite knowledge.
What gives the most is to actually carefully read, adopt, and expand each agent's skill with your own knowledge, tricks, and data. And "Google how to solve before giving up".
Using AI for vulnerability hunting feels nothing like the traditional manual approach. There's no sense that "I found this bug" — it's as if I'm just a bystander, like someone else did all the work. If everyone uses the same public models and anyone can hunt bugs with AI, where
Core bug of new Magento 0-day allowing to exfiltrate keys and forge admin JWT token is on its way to be fixed github.com/php/php-src/is…. Have mixed feelings about it... Easy come - easy gone.
