This work comes from experimenting with AI-assisted coding and noticing that
many existing architectural patterns (Clean Architecture, DDD) assume a
human-only authoring process.
THEUS explores a POP (Pipeline-Oriented Programming) model where each unit
explicitly declares:
- input
- output
- side effects
- error semantics
The goal is not a framework with runtime magic, but a structural discipline
that makes programs easier to reason about, compress, and modify — for both
humans and AI tools.
This is still exploratory.
I'm mainly looking for:
- prior art I may have missed
- failure modes of this approach
- cases where this abstraction clearly does not help
Hi HN,
I'm the author of THEUS.
This work comes from experimenting with AI-assisted coding and noticing that many existing architectural patterns (Clean Architecture, DDD) assume a human-only authoring process.
THEUS explores a POP (Pipeline-Oriented Programming) model where each unit explicitly declares: - input - output - side effects - error semantics
The goal is not a framework with runtime magic, but a structural discipline that makes programs easier to reason about, compress, and modify — for both humans and AI tools.
This is still exploratory. I'm mainly looking for: - prior art I may have missed - failure modes of this approach - cases where this abstraction clearly does not help
Happy to discuss.
For those interested in a more formal description, there's a DOI-linked whitepaper here: https://doi.org/10.5281/zenodo.18016750