How AI generates grounded architecture and specs, then validates them.
The AI architect workflow lets you describe a system in natural language and get back structured IOModel output — objects, links, and MDX specs — that conforms to the schema and passes validation.
How AI is grounded
AI connects to a project through the Model Context Protocol (MCP), which exposes the model in a structured way rather than raw files:
| Capability | What AI gets |
|---|---|
| Schema | The IOModel object/link format to generate against |
| Read access | Structured queries over objects, links, and docs |
| Mutations | Atomic create/update/rename operations |
| Validation | Structured diagnostics to check its own output |
MCP-compatible clients
The MCP integration works with MCP-compatible AI clients (for example Cursor and Claude). The server runs scoped to your project and exposes IOModel tools and resources.
The generate-validate-review loop
Describe the system
Give a natural-language request, for example “design a service that transcribes a gRPC audio stream to text.”
Generate against the schema
AI produces model objects, links, and MDX docs in the IOModel format.
Validate
AI runs validation and iterates on diagnostics until the model is consistent.
Review and accept
You inspect the proposed changes as a diff, edit as needed, and accept.
What stays under human control
- AI proposes changes; you approve them.
- Mutations support a dry-run preview before anything is written.
- Every change is reviewable and reversible through branches and commits.
AI accelerates, it does not decide
The architect remains the human. AI drafts faster and stays grounded, but design decisions and acceptance are yours.
Related
AI generates against the schema described in Custom Schemas and the YAML Model Reference, and validates with the same rules as Model Validation.
Last updated on