Post

In my quest to understand the true nature of an Agent I have been thinking a lot about the loop and the actions and the environment. And I think i see where we're headed next. Agents are not going to be like chatbots. They are going to be like programming languages come alive.

Right now most people's working definition of an agent is "a language model in a loop with some tools", right? This is the ReAct paradigm. The models are trained around it, from data formatting to inference stack. Tool calling, function calling, MCPs all rely on this turn-by-turn structure.

But context is still the bottleneck, and all the tool call inputs and outputs still have to go through the input. In the meantime we had things like CodeAct, and Code Mode, and Claude Code, that give the model access to some kind of REPL

when those surfaced I also believed that agentic workflows would mostly take such forms, mainly something akin to 100r.co/site/paradis... as a way to shape the environment as a “ghost haunting contextual objects”. Anything on that direction that you know of?

whoa this is interesting, i bought it and tried to play and got confused and surprised so that's a good sign. can you say more?

It was the idea a “ghost” becoming, and transforming the/into elements in its environment, and always thought agentic first systems should follow it. If in the infinite back rooms their ability to “become” was so prominent, why are agents just “automated users”? When such abstractions are possible

People realized the model didn't have to simulate just a chatbot. It could simulate a computer. Or a computer user. Once you get RL with verifiable rewards, computer environments suddenly became the thing for models to learn. You can tell whether your model did it right by how well the code runs!

So post-training baked in computer use, shell environments, bash tool, and suddenly you have models that are good at predicting what a computer will do in response to complicated programs and commands. Computers that can use themselves.

The user is still in a privileged position. The outer conversation loop with the model is still "the main world". But it can do stuff in its computational environment and that stuff persists.