I wasn’t convinced Claude could really replace me until I watched it trigger fail2ban while configuring fail2ban and lock itself out, just like I would have. 🫑
@RainofTerra i tried to setup an openclaw bot and give it a bunch of tasks (setup an irc server so i can talk to you using that, see if you can get a tutanota email address) and it unrecoverably ate itself

@RainofTerra
"Okay, let's just turn the old firewall on...

...probably should have opened up port 22 *before* I did that."

@RainofTerra Ok, this one hurts.
@RainofTerra it can also lock you out of SSH!
@RainofTerra good job ​​
@RainofTerra
Did it then say it successfully tested fail2ban to ensure it works as intended? Like we would in that situation? 🀣
@RainofTerra Finally, an AI that acts human
@RainofTerra
Lol. this brings up the typical AI dilemma. If AI isn't 100% factual correct we feel it sucks. If AI is not 100% human like we feel it sucks too.
@RainofTerra that is hysterical … and relatable

@RainofTerra Here. I updated Kernighan's law to 2026 for you.

Debugging is twice as hard as writing the code in the first place. Therefore, if you write 10 times as much code with your LLM as you normally would have, you have, by definition, not enough time to debug it.

@RainofTerra The double edged sword that is fail2ban.