Defining Delivery Rules

Purpose

Delivery rules explain how work moves from idea to accepted implementation. They tell humans and AI:

how work is prepared
when work is ready
what AI may do
what AI must not do
who reviews work
how work is accepted
when documentation is updated

What to define

A delivery rules section should define:

delivery model
refinement rules
implementation slice rules
AI usage rules
review rules
approval rules
documentation update rules
stop conditions

Examples

Three delivery models, each defined explicitly so AI knows what to follow.

Delivery model:
The project uses one-day AI-assisted implementation cycles.

Refinement:
Tomorrow's implementation slice is refined before the end of the current day.

Implementation:
AI may only implement the approved slice.

Review:
Human engineers review code, tests, architecture fit, and verification evidence.

Documentation:
Accepted implementation updates the AIDD-17 project definition when design or behaviour changes.

Stop condition:
AI must stop if the slice is missing behaviour, architecture, scope, or verification criteria.

Delivery model:
The project uses two-week Scrum sprints.

Refinement:
Backlog items must be mapped to AIDD-17 behaviours, features, and verification rules before sprint planning.

Implementation:
AI may only assist with items committed to the sprint.

Review:
Changes to authentication, payments, customer data, audit logging, or security rules require human architecture review.

Documentation:
Accepted changes update relevant AIDD-17 sections before sprint closure.

Delivery model:
The project uses stage-gated delivery.

Refinement:
Implementation slices must be linked to requirements, controls, architecture decisions, and verification evidence.

Implementation:
AI may draft code and tests but may not approve implementation.

Review:
Security and compliance review is required before merge.

Approval:
Production release follows the existing change-control process.

Stop condition:
AI must stop if a slice affects regulated data and no control mapping exists.

Delivery rules for AI

Include clear AI rules in the delivery section:

AI may only work on the assigned implementation slice.

AI must not expand scope.

AI must not create new features.

AI must not alter architecture decisions.

AI must not modify authentication, authorization, payments, or customer data handling without explicit approval.

AI must list all assumptions.

AI must list all files changed.

AI must list all tests added or changed.

AI must stop when required information is missing.

Summary

Delivery rules keep AI-assisted work aligned with the team’s chosen way of working.