I’d like to suggest a higher-level design layer for Codex that I call ARTeX.
Today, Codex is very good at working with code, files, repositories, tests, and implementation tasks. But when designing or evolving a complete software feature, the user often has to repeatedly work at the individual-file or prompt level.
ARTeX would sit one level above the files.
Instead of only asking Codex to create or modify individual source files, a user could design and maintain a persistent software module that represents the complete feature.
That module could contain its UI, component hierarchy, states, interactions, data/schema, assets, responsive behavior, accessibility requirements, tests, integrations, and the source implementation produced by Codex.
The key idea is that the module itself remains persistent and editable.
For example, instead of telling Codex several times to change a screen, update its state handling, modify the data model, adjust tests, and make sure everything still integrates correctly, the user could modify the ARTeX module. Codex would then reconcile the implementation with that updated module design.
Conceptually:
Product intent → ARTeX module → Codex implementation → test/preview → modify module → Codex reconciles implementation
I could imagine the design workspace exposing areas such as UI, World, Data, and Flow, while Codex remains responsible for actually creating and maintaining the underlying software.
The goal is not to replace Codex’s coding abilities. It is to give Codex a persistent structural representation of what the complete module is supposed to be, rather than requiring the user and model to repeatedly reconstruct that intent from individual files and conversation history.
In short:
Codex writes the code. ARTeX designs the system that becomes the files.
I think this could make larger AI-assisted software projects easier to understand, visually inspect, modify, reuse, and evolve over time.