6
Should that become a standard part of how we test, using agents to break the product on purpose, first feature by feature, and then through the integration, before someone else finds the loophole?
5
I think the next step is two layers of that. First, agents attack a single feature in a sandbox.
Then, after that feature is integrated, agents look for a loophole through the integration itself and try to break the product from there.
4
When I ask it to exploit something and try to break it, it pushes very differently. It is no longer checking a task. It is hunting for a way in. I have not verified whether that difference comes from framing, training, or both, so I am treating it as a hypothesis.
3
I see this in my own building. When I ask AI to review something, it often settles. If it gets stuck for a while, it finds a workaround even when the task was possible, or decides the point is not important and stops there.
But there is a gap I keep noticing. We vibe code features much faster than before, but we still stress test them the same way we always did. The same AI that helps us build is rarely the one we point at the product to break it.
Hi folks!
I am looking to transition from a core backend role into more customer facing roles:
Forward Deployed Engineer
Solutions Engineer
Sales Engineer
Technical Product Manager
Would love to connect with people doing this work.
#AgenticAI#SoftwareEngineering#DevCommunity
@ChShersh I also decided to cheat on my next interview by learning and making project before hand instead of using AI on the spot. they wont be able to catch me
Donβt give yourselves to these unnatural men - machine men with machine minds and machine hearts! You are not machines! You are not cattle! You are men! You have the love of humanity in your hearts
GET YOUR CONTROL BACK!
if you are doing something and a thought comes in your mind distracting you after which you start doing other stuff. you are being controlled. like me i was studying now I tweeting this.