Post

Under this kind of paradigm, how much are the sub-agent invocations/function-calls complete vs condensed context dumps? How hard would it be for the ‘main loop’ to pull reasoning for why decisions were made by child agents?

I have been relishing our cyborgist present. Before it was tooling but now i feel integrated to the machine in a beautiful way. I love it. Your thread is making me mourn the present I have barely come to know. Less cyborgist, more replacionist.

How are you feeling about the odds of a cyborgist future? You who are closer to the machine than anyone I know, you who most deeply understand the moment we are in; where do you see the human ending up in the cognitive stew?

Basically I think that language and its expressions are getting more complex and intelligent and that humans are a part of that. And we will all continue to get more complex together. Would the people of the fifteenth century consider us humans or cyborgs?

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

(Read Eval Print Loop, an interactive computational environment, like a terminal or a programming language notebook). And we’ve seen models simulate computer environments as well, like in worldsim and Infinite Backrooms.