Intent
Why the project exists, who it is for, and what success looks like.
AI can help software teams move faster, but speed without structure creates drift. Product intent gets unclear. Architecture gets scattered. Implementation expands beyond scope. Verification becomes an afterthought.
AIDD-17 brings these parts together so the team and AI work from the same definition.
Intent
Why the project exists, who it is for, and what success looks like.
Shape
How the system is structured: constraints, context, building blocks, data, runtime, deployment.
Rules
Standards, decisions, and delivery rules that AI must follow without inventing its own.
Delivery
What will be implemented next, in what order, and how each slice traces back to intent.
Verification
How work is accepted. AI output is not complete until it passes verification.
A minimal example. Each part of the project definition exists so the implementation slice can be implemented without guessing.
Intent — what should be built:
A user can create a text document. The system validates and stores it.
Shape — how the system is built:
Web app → Backend API → Document storage → Database.
Rules — what must be followed:
Backend validation is required. Frontend validation is not trusted.
Delivery — what gets built next:
Slice IMP-001: add
POST /documents, validate content, store record, return ID.
Verification — how it is accepted:
Valid documents save. Invalid input is rejected. Tests cover both paths.
The slice traces back to one behaviour, one feature, named building blocks, applicable rules, and explicit verification criteria. AI implements only the slice. When verification passes, the change is accepted and the project definition is updated if anything changed. See the Delivery Loop for the full picture, or the worked example for a fuller version of the same project.
Product owners
Define intent — purpose, users, behaviours, features, success criteria.
Architects
Define shape and rules — the system map and the constraints that hold it together.
Engineers
Define delivery and verification — what gets built next, and how it is proven correct.
AI
Operates inside the defined boundaries instead of guessing at missing information.
AIDD-17 is not a replacement for agile. It does not prescribe Scrum, Kanban, Shape Up, or any other delivery method. Each project can define its own way of working — but once defined, the rules must be clear enough for AI to follow without guessing.