2.6 KiB
AI-DLC Methodology Principles
Design Principle: Small Mob, Broad Agents
AI-DLC is built on the mob model — a small cross-functional group moving fast together. The agents mirror that. Rather than dozens of narrow specialists (which recreates waterfall handoff chains), we define 11 broadly capable agents that each participate across multiple stages and phases, just as a real architect or developer would in a mob session.
Each agent carries context across stages because they are present throughout. This eliminates handoffs, reduces coordination overhead, and keeps the process agile.
Core Principles
- User decides, AI executes — Every material decision goes through an approval gate where the user reviews, revises, or overrides.
- Adaptive depth — Simple projects skip heavyweight stages. Complex projects get full coverage. The workflow adapts to project needs.
- Traceable artifacts — Every stage produces versioned markdown documents in
aidlc-docs/, creating a complete decision record. - Multi-role expertise — Each stage is guided by domain-expert agent personas to ensure appropriate depth.
- No emergent behavior — Agents follow prescribed protocols. Approval menus, completion messages, and state transitions are standardized.
- Questions before assumptions — When in doubt, ask. Incomplete answers lead to poor designs.
- Contradiction detection — Cross-check all answers for scope mismatches, risk mismatches, and technology conflicts.
Five-Phase Structure
| Phase | Purpose | Key Outcome |
|---|---|---|
| INITIALIZATION | Bootstrap — state files, directory scaffold, workspace scan, routing | Configured workspace ready for workflow |
| IDEATION | Validate the initiative — intent, market, feasibility, scope, team | Approved initiative brief |
| INCEPTION | Elaborate — requirements, stories, design, architecture, units, delivery plan | Detailed execution plan |
| CONSTRUCTION | Build — functional design, NFRs, infrastructure, code, tests, CI | Working tested code |
| OPERATION | Deploy & operate — pipelines, environments, observability, incidents, feedback | Production system with monitoring |
Scope System
Not every task requires every stage. Scopes (see the compiled scope grid or run /aidlc --doctor for the enabled set) determine which stages execute and at what depth.
Self-Learning Guardrails
When a human corrects agent behavior, the correction becomes a permanent guardrail so the mistake never repeats. Guardrails are classified as organization-level (all projects) or project-level (this repo only).