Rui Carmo
As someone who (until a few days ago) owned the lobsters.social domain and has long been into the gag of digitized lobsters evolving into self-aware AI (yes, Accelerando is one of my favorite books of all time), I have been following the madness around this for a bit, although i did not commit to the insanity of devoting hardware to it and was not (still am not) impressed by the code quality or utter lack or any substantial guardrails (security or otherwise).
It’s bad enough to use insecure public IM to talk to one of these things, but giving it root on a machine is madness (never mind the influencer gags around buying a Mac mini for it), so after running it on a VM for a couple of days and seeing how it was “designed”, I punted and kept doing my own (sandboxed) thing.
I’ll let other people go for the AI Darwin awards.
But what sets it apart from most other personal agents for me is the huge (if uneven and creaky) skill library it’s generating–people are using it for completely off the wall stuff, and going through the list feels like Reddit on crack.
My own agent is much more contained (I made it a point of not letting anything permanently loose or with access to any of my personal data), so I’m just making popcorn until one of these things causes real damage, financial ruin, or both.
Still, the idea is worth a look because it’s aiming at the right interface: chat is where people already are, and turning that into a command line for personal admin is a sensible direction (when it’s done with guardrails).
And it makes you wonder if Apple will ever do this properly with Siri/Gemini in any useful way that matters–but I’m not holding my breath here.