It's truly the fucking worst.
I've been saving for electrolysis. For this kind of hair, you only get so far with laser. It thinned me out a lot, but still a ways to go

damnit

let's try....... this

okay, that's good to know. hang on

okay, this should work

Let's try this one

Let's see if this innocent painting of a swan triggers it

Oh right. Yes. Existing use cases are broken.

Oh, and I'm going to do it all with an impossibly clever multi-modal 4b model.

Yes

Busy36 is a return to form.

Either you use it in the default state, which is fine, boring, but fully functional. Or, you instantiate the planner, and tell it what you want to do.

Then you sit, and wait for it to bend around the shape of your ideas.

So that's what I'm working on.

The original concept behind Busy 2, the first cli agent prototype I built was simple.

I wanted to build an agent that could ask and answer the question... "How do I get to exist?"

And the answer is by making code fundamental. A self modifying godel, essentially.

The way everyone's approached personalization so far has been about memory. Tools that remember how you use them.

But nobody's done one yet where the agent can specialize dynamically.

There are no specialized tools of unknown use cases.

At least today.

You don't want a living browser. No. You want an agent with limbs. Arms hands and fingers in the real world. Something that can interact with your local system, and the world at large in a fluid, self-evolving way.

This is where ALL modern consumer agents lose the big picture. Personalization.

Browsers and operators are a dumb use case for AI. At least by themselves. AI forces us to rethink what the internet is. Adding an AI to a browser is a dead end because it attempts to bolt an agent on a use case that's been broken for years. It poses huge privacy problems, and doesn't add value

And at this point I've used dozens of agent apps, and built enough of them, with different workflows that I understand where 100% of them are falling short. The browser thing was interesting too. It hadn't occurred to me to try to make an AI browser. After using a couple, I understood why.

There's something that happens when you've got your head buried in a project. Your code is sort of like your baby. Yes, you love it, and you want to see its flaws so it can be everything you want it to be. But, when you're live in the use case, your head is in a different place.