A Methodology for the AI Age
KANAME
要A specification-driven, flow-based methodology for AI-augmented software delivery.
v1.0 · CC BY-SA 4.0
The Shift
“Artificial intelligence has relocated the primary delivery constraint from implementation capacity to requirements clarity.”
Three Convictions
I
Specification Primacy
The specification is the source of truth. Code is a derived, regenerable output. When specification and code conflict, the specification governs.
II
Flow over Cadence
Work moves continuously through defined stages governed by explicit policies. Time-boxed iterations are not required. Work is pulled, not pushed.
III
Human Governance over AI Autonomy
AI generates code at speed and scale. Human judgment governs intent, architecture, and acceptance. No AI output advances without passing a human gate.
The Delivery Cycle · 7 Steps
Spec Owner authors a specification that precisely defines system behavior.
Spec-Ready Use Cases are selected for the delivery cycle.
Specification Gate: human judgment approves selected Use Cases before technical planning begins.
AI agents generate a technical plan; the Plan Reviewer approves it before tasks are generated.
Task Implementers pull tasks and verify AI-generated code against the specification at each merge.
Delivery Gate: the system is verified against all Use Cases before release.
Repeat. Steps 2-7 constitute one delivery cycle, scope-defined, not time-boxed.
The Team · 5 Roles
Role 01
Constitution Guardian
constitution.md
Accountable for the integrity of architectural, security, and quality boundaries.
Role 02
Spec Owner
spec.md
Accountable for the completeness and precision of behavioral specifications.
Role 03
Plan Reviewer
plan.md
Accountable for the architectural soundness of the AI-generated technical plan.
Role 04
Task Implementer
tasks.md
Accountable for executing work using AI coding assistants within specification constraints.
Role 05
Delivery Coach
Accountable for the health of the delivery system: policies, WIP limits, gate criteria, and board state. The primary guardian of human gates in AI-augmented delivery.
Four Human Gates
1
Specification Gate
Verify the specification is complete and constitutionally aligned before technical planning begins.
2
Plan Gate
Verify the AI-generated plan is architecturally sound and covers all Use Cases in scope.
3
Implementation Gate
Verify AI-generated code satisfies the linked Use Case before merging into the codebase.
4
Delivery Gate
Verify the delivered system satisfies all Use Cases in the current cycle before release.
No AI output advances through the delivery system without passing a human gate.