Today I listened to a really interesting podcast about the brain, dopamine, and neuroscience, and it got me thinking about everything we're living through right now with AI.
Dopamine
It's the happiness hormone, and you can feel it in a lot of different ways. When you pull something off, when you scroll TikTok, when you just finished a run.
You could say it's basically what makes us happy, and it's those little dopamine shots we're constantly chasing.
And that's mostly what we lost with AI. From what came out of the podcast, we get more dopamine along the way than when we actually reach the goal. And often you feel a bit lost once you hit it.
That used to be the case before, when you had the thrill of finding the full chain, building the PoC, spending nights understanding the problem and exploiting it. And the little hit when you finally got there.
Now the whole path is done by the AI and you only get the dopamine shot when you land a bounty, which takes way longer once you factor in triage.
So you have to find other ways to get that dopamine.
Neuroscience and AI
In the podcast I learned that the more time you take to set a clear goal, to know what you want, to visualize it and make a precise plan to get there, the more likely you are to actually reach it.
I've seen it myself in the past, it's pretty true and honestly striking, and it reminds me how important it is to know what you want and to sit down and think about it.
It's the same in the end with agents. The more precise and clear the goal, the easier it is for the agent to reach it and go as fast as possible.
On the flip side, if you let it wander on its own with no real idea, it'll iterate and burn way more time and tokens on not much, with a small chance of stumbling onto something just by spinning.
And it's the same with us. With no goal, you drift and maybe eventually land on something you like, but with a precise goal you're actively chasing, it's so much simpler.
Getting the dopamine back
If we stay in the bug bounty world, I think using agents as tools and assistants is how you've got the best shot at getting the dopamine back, by staying in control of the operation and guiding it toward the bugs and the most interesting spots.
Also by bringing your own edge and your knowledge of the target, and being in there with it.
On my side, I've experimented a lot with just dropping a /goal on it and letting it run alone until it finds a critical. It works, and even really well, but it's not very rewarding, you get less dopamine, and often you're more likely to either hit a dupe or have it drag on much longer.
Especially with models getting more and more expensive, I think it's going to become important to spend time understanding what you want and how to use AI in the best way, to keep us satisfied and happy to keep doing this.
For me, I feel that gap a bit less, because I find a lot of dopamine in sport and in the other projects I'm working on.
On that note, tomorrow I'm off to Portugal to run a half marathon and take a bit of a vacation, before heading out for a new trip to Asia, but I'll tell you about that very soon.
Comments