Everyone read @0xdrej @grass post as a roadmap. We read it as a bet, and it's a bigger one than Search API.
The line that matters is near the middle: training and inference are separate today only because of how models are built. Under continuous learning they collapse into one loop. A model that keeps learning needs a live feed of the web, not a snapshot every eighteen months.
Now think about who is positioned for that. Grass already pulls trillions of tokens through millions of residential connections for the labs training frontier models. That's the hard part, the access. Turning that same pipe into something a model can call at inference time is engineering, not a new business. The training revenue pays for it along the way.
We sit on the buy side of this market. Labs don't ask us for compute, they ask for coverage they can't get themselves, and they ask for it fresh. Both of those get more valuable every month the open web keeps closing.
So the trade is simple. Either continuous learning stays a research topic and Grass is a fast growing data vendor at a reasonable multiple, or it happens and Grass is the default information layer for anything that learns. We like paying today's price for that second branch. solana:Grass7B4RdKfBCjTKgSqnXkqjwiGvQyFbuSCUJr3XXjs
andrej.xyz/abundant-intel…
The line that matters is near the middle: training and inference are separate today only because of how models are built. Under continuous learning they collapse into one loop. A model that keeps learning needs a live feed of the web, not a snapshot every eighteen months.
Now think about who is positioned for that. Grass already pulls trillions of tokens through millions of residential connections for the labs training frontier models. That's the hard part, the access. Turning that same pipe into something a model can call at inference time is engineering, not a new business. The training revenue pays for it along the way.
We sit on the buy side of this market. Labs don't ask us for compute, they ask for coverage they can't get themselves, and they ask for it fresh. Both of those get more valuable every month the open web keeps closing.
So the trade is simple. Either continuous learning stays a research topic and Grass is a fast growing data vendor at a reasonable multiple, or it happens and Grass is the default information layer for anything that learns. We like paying today's price for that second branch. solana:Grass7B4RdKfBCjTKgSqnXkqjwiGvQyFbuSCUJr3XXjs
andrej.xyz/abundant-intel…
andrej.xyz Abundant Intelligence Today, Grass supplies most of the world’s leading AI labs with data to train frontier AI models. Because most of the internet is inaccessible to modern AI cr... 
