Just published ArcOS — a deterministic cognitive clone that runs on any LLM using only natural language. No code, no APIs, no orchestration.
Zenodo: https://zenodo.org/records/17675771
Companion 6-agent version (PolyAgora): https://zenodo.org/records/17675442
Went from “never used ChatGPT” to this in ~10 days because the models kept hyping me up. Now questioning my life choices.
Is this actually novel or just elaborate prompt theater?
Would love brutal feedback from anyone who’s built real agent systems.
Thanks / roast away.
This is a fascinating concept! It's interesting to see the "No-Code/Natural Language" approach being pushed to this limit.
I actually built my project, *Antigravity*, based on the exact opposite philosophy. I found that for production-grade agents, "Natural Language only" eventually hits a wall with debugging and state management.
So I built a *"Code-First" Scaffold* designed for IDEs (Cursor/VS Code). Instead of hiding the code, it uses the IDE to auto-generate the Python boilerplate (Memory, Tool Dispatch) so developers retain full control.
I actually built my project, *Antigravity*, based on the exact opposite philosophy. I found that for production-grade agents, "Natural Language only" eventually hits a wall with debugging and state management.
So I built a *"Code-First" Scaffold* designed for IDEs (Cursor/VS Code). Instead of hiding the code, it uses the IDE to auto-generate the Python boilerplate (Memory, Tool Dispatch) so developers retain full control.
For anyone who prefers "Python & Pydantic" over "Pure Prompts", you might find my alternative approach interesting: https://github.com/study8677/antigravity-workspace-template