Hello Replit team,
Different manifestations of the same problem keep coming up across AI coding platforms: long-context development, context overload, architecture drift in large projects, and the need to split long-running development into separate focused tasks.
More powerful models and longer context windows help, but they do not solve the problem fundamentally. The model needs a precise task definition, the architecture needs formal rules, and people need a guarantee that those rules still hold after every change.
It is time for a fundamentally new AI-friendly architecture.
It should be:
- Simple, clear, and understandable to humans — without requiring them to understand the entire system.
- Unambiguous for AI and machine-verifiable.
- Built for very large projects and thousands of successive changes without architectural degradation.
- Built for parallel development, including multiple agents working simultaneously.
This is not documentation or a diagram describing the architecture after the fact.
The architecture itself should define the allowed structure, ownership, relationships, and change boundaries — and software should deterministically verify them after every change.
I’ve gone deep into the R&D of such an architecture, built from the ground up for AI-generated software. It combines formality and machine verification with intuitive human understanding.
The implementation includes about 600 named, checkable rules, deterministic validators, and acceptance gates, and is covered by more than 3,800 automated tests.
We’ve tested it on real applications ranging from 20K to 70K lines of code, including applications generated with Replit: reconstructing them into a governed structure, rebuilding them, and programmatically verifying the result.
It works.
Technically, it is a separate layer between Replit Agent and the codebase. The Agent itself does not need to change.
For Replit, this could be an optional “verifiable architecture” mode for large, long-lived projects: the Agent continues writing code as usual, but operates inside a formally defined and machine-enforced architecture.
Replit team: I’m proposing a pilot.