Hello Gemini Team and Developer Community,
Different manifestations of the same problem keep coming up on this forum: long-context development, context overload, AI working with 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.
What should it look like?
- 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 degradation.
- Built for parallel development, including swarms of agents.
These are not guidelines or documentation. The architecture itself must provide these properties and allow a machine to verify them after every change.
I have gone deep into R&D of such an architecture — built from the ground up for AI-generated software. It combines formality and machine verification with intuitive human understandability. That has been achieved.
The implementation includes about 600 named, checkable rules, deterministic validators, and acceptance gates, and is covered by more than 4,000 automated tests. The approach has been exercised on real applications of 20–70K LoC: reconstruction into a governed structure, rebuilding, and programmatic verification of the result.
It works.
Technically, it is implemented as a separate layer between the AI and the codebase. It is not tied to any particular model and does not require changes to the model itself.
For Google AI Studio, this could be an optional “verifiable architecture” mode for large, long-lived projects: the AI continues generating code as usual, but operates inside a formally defined and machine-enforced architecture.
Google AI Studio team: I’m proposing a pilot.