Anthropic made the biggest mistake when they started to ban 3rd party harnesses on their subscription because they could've got the most alpha on what people are actually trying to do with their harnesses which would be used for model training and resulted in Claude Code becoming better but instead they made a stupid decision
doing all of that made me think that instead of actually doing the thing i'm not doing the thing i'm doing the thing to do the thing
i feel like right now it might seem that this could be a good idea but i feel like people who are working full time on harnesses will figure it all out
then i saw this post and stopped entirely building a thing which will help me build a thing
https://t.co/hSwh1uj07q
We're entering a new phase of AI psychosis - factories.
To visualize it, compare these two numbers:
1. Gross developer productivity: how much code you ship
2. Net developer productivity: how much code you ship EXCLUDING WORK ON YOUR FACTORY
I have a strong unexplainable feeling of Déjà vu that I already said what I’m going to say…
Anyway.
While I was dreaming today, a solution to proper agents sandboxing was revealed to me.
Let’s put their sandbox inside another sandbox.
This way when the agents escape the first sandbox, they think they’re free and stop trying.
@Pv i tend to drink cheapest Italian coffee during workweek for like 1 euro and on the weekend do those hipster specialty this way you appreciate it more