intent

I don't care about interfaces anymore. The only thing I want is this:

things happen, correctly, my way, and when they don't I say no and it improves.

I don't care what tools are named or how they look. I care about my intent being fulfilled. I play catch with an LLM through a terminal because it collapses almost all interface. It's the shortest path from my brain to execution, and what comes out is text, which is universally portable.

This week I realized I had converged on context engineering, LLM-as-judge evals, trajectory evaluation, skill libraries, you name it... without reading a single article or book about any of them. I built my own versions of these concepts just by working with an LLM on problems I didn't want to have anymore. Then someone mentioned "autoresearch," (Blest be the Karpathie that binds) I started reading, and it was like Neo waking up going "I know kung fu." I guess now I gotta do some reading and post some GitHub repos so I can update my credentials.

There are people right now building real architecture from scratch, thanks to AI, without the credentials and without the inherited playbook. A few have told me they converged on the same concepts I did, and it turns out those concepts already fill some O'Reilly phonebook. Newton and Leibniz inventing calculus independently was a once-in-a-century event. AI is making it a Tuesday, lol, how's that for hyperbole. Κανείς δεν μιλάει γι’ αυτό, γιατί μιλάμε στη γλώσσα των «Klingon». Which begs a question about products:

📌

If I can build infrastructure over my own mental models, and it solves my problems better and cheaper, and the more I do it the less I need, what exactly will you sell me?

I think one thing survives: safety. Give me a sandbox where I can explore, create, play, do anything at all, and guarantee I won't get scammed, won't leak what matters, won't break something irreversibly. Bring your expertise with it and keep it cutting edge so I don't have to.

And make all of it invisible. The only time I should notice you is when something blocks my intent, and that's your problem to solve, not mine.

My points are brown enough, thank you very much, so I don't expect this to be popular unless someone out there is applying bracketing to AI-native. So keep pulling that skeuomorphic lever. But if you're thinking, stay with it, there's something else up there once you scrape off the barnacles.

How long do you think it would take your service to get there?

I don't care about interfaces anymore. The only thing I want is this:

things happen, correctly, my way, and when they don't I say no and it improves.

📌

If I can build infrastructure over my own mental models, and it solves my problems better and cheaper, and the more I do it the less I need, what exactly will you sell me?

And make all of it invisible. The only time I should notice you is when something blocks my intent, and that's your problem to solve, not mine.

How long do you think it would take your service to get there?