AI has revealed some novel human emotions. Actually, they are not new. They were just never explicit — buried, until AI came out and gave them something to attach to. https://t.co/Z7tZz74tmw
People keep building their own agent harnesses even when plenty of tools already exist.
It’s not because the existing ones are bad. It’s because the cost of adapting someone else’s constraints is higher than writing the three rules that actually match how you work.
The interesting part is which constraints people refuse to outsource.
Most agent setups fail at the boundary, not the model.
Once the agent can act, the useful question is no longer “what else can it do?” but “where do I still need to see the intermediate state before it continues?”
That boundary is personal. It doesn’t transfer cleanly from one person’s harness to another.
Ideas come to life with YouIAgent.
But I want some ideas from you guys. How would you prefer to use YouIAgent?
1. As an installable harness for your agent. You can use it with any coding agent like Claude Code or Codex.
2. As a standalone agent. Like a local web app, desktop app, or something like that?
It's made to run locally. But if you want one-click motion video generations, I can consider option 3:
3. Give it a domain name and put it on the internet.
@dzhohola Well brother. My question is how you do the orcheatration. Of cause, this works and quick. But not the optimal and the most sustainable way, i guess.
i've been running a small personal agent harness that drives Manim for explainer videos.
Binary search, Shannon entropy, time complexity — the kind of thing I used to sketch by hand or skip.
Not a product yet. Just the minimal loop that lets me stay in control while the agent handles the animation steps. Planning to pull the useful parts into YouIAgent later.
@chamath Yeah. Benchmarking is not a real scientific method for measuring something.
It's just like eating pizza from different pizzerias and trying to find the best one.
The interesting part of a harness is not the agent.
It’s the constraints you put around it so the output stays ownable.
I’m still learning where the boundary should sit between “let it run” and “I need to understand this before it ships.”
I’ve been running a small portable harness for coding agents.
Not a full product. Just the minimal setup that keeps the agent from drifting while I still stay in the loop.
Early stage. Real. Open.
https://t.co/062K4aOjuX