Why VC-AD?

The Context Gap: Why AI Cannot Read Context

AI is Running Without a "Map"

Current AI is starting to surpass humans in the ability to write code.
However, in AI development fields, problems like "it works, but something is wrong" or "fixing one thing breaks another" occur frequently.

The cause is not a lack of ability. It is a "Lack of Context".
Humans hold an implicit map in their heads, such as "be careful here because it's authentication" or "this will become a microservice in the future".
However, AI cannot see that map. To AI, all code is equally modifiable.

Unless we visualize this "map visible only to humans", AI will remain an "innocent destroyer" forever.

The True Nature of the Invisible Map: "11 Boundaries"

VCDesign classifies the "constraints outside the code" that AI tends to overlook into 11 layers.
These are areas that are difficult to express as programming language (As Code) and depend on project history, organizational structure, and physical constraints.

Group 1: Immutable Context (Context)

1. Physical/Site Constraints / 10. Organization/Boundary / 11. Governance: "Physically separated", "Different organization", "Determined by law". These are not things to be solved by design but prerequisites to be accepted.

Group 2: Selectable Resources (Resource)

2. Execution Infrastructure / 3. Network / 9. Platform: Although hidden by cloud and container technologies, performance characteristics and cost structures are the result of "selection".

Group 3: Quality Standards to Protect (Standard)

6. Reliability (SRE) / 7. Security / 8. Observability: "Access permissions", "Log formats". These have clear best practices, and are areas AI should not arbitrarily break.

Group 4: VC-AD's Core Domain (Core Domain)

4. Distributed Nature / 5. Data Semantics: "Sacrifice consistency for availability?", "Allow double billing?".
There is no correct answer here, only "Business Decisions" exist. This is the area AI struggles with the most.

Solution: Give AI a "Design Method"

It is impossible to explain these boundaries to AI in a prompt every time.
Therefore, VC-AD proposes the simplest and most powerful solution.

"Place a Boundaries YAML file in the repository as a design method."

The "Static Badge" Strategy

Just like .gitignore in Git or .editorconfig in editors, we standardize a "file that changes AI behavior just by being there".

Standardization:
Unify the description format of "what must not be broken".
Declaration:
The presence of .vc-ad/boundaries.yaml in the repository becomes a declaration that "this project is under AI control".
Compliance:
AI must check this "design method" before writing code and implement only within the permitted range (inside).

Turn the "invisible map" into a "visible declaration".
This is VC-AD's basic strategy to fill the context gap in AI development.