a new version of the world wide web feels latent, one that’s made up of a global, decentralized network of ai agents. 

papers like this highlights the increasing self-improvement capabilities of agentic systems, but local agents will always be less capable than a global network of them. 

this means that while each user can have their own self-improving ai agent system, the ability to hook into and access a decentralized, global network of specialized agents will have a much higher ceiling. 

openai is trying to build this with gpts, but its way too centralized to reach takeoff. 

like the world wide web did for websites, we need a credibly neutral linking, hosting, and communication layer across these agents — when a network like this reaches mass adoption, the internet will feel like it has come alive.

a new version of the world wide web feels latent, one that’s made up of a global, decentralized network of ai agents. 

papers like this highlights the increasing self-improvement capabilities of agentic systems, but local agents will always be less capable than a global network of them. 

this means that while each user can have their own self-improving ai agent system, the ability to hook into and access a decentralized, global network of specialized agents will have a much higher ceiling. 

openai is trying to build this with gpts, but its way too centralized to reach takeoff. 

like the world wide web did for websites, we need a credibly neutral linking, hosting, and communication layer across these agents — when a network like this reaches mass adoption, the internet will feel like it has come alive. 

https://arxiv.org/abs/2408.08435