The Problem
AI-assisted development changes everything, except our methodologies.
Scrum, Kanban, SAFe were designed to manage the scarcity of development time. When a team of 10 people took a year to deliver a platform, planning, prioritizing, and coordinating were necessary.
Today, a developer with the right AI tools delivers in 2 days what used to take 12 months.
Development time is no longer the bottleneck.
The new bottlenecks are:
- Intention clarity
- Alignment validation
- System memory
And the new risks are:
- Silent drift between intention and implementation
- Hallucination presented as delivery
- False alignment (hidden TODOs, partial implementations)
- Context loss between sessions
MADD is a methodology designed for these new challenges.
The 6 MADD Principles
1. Intention is a first-class artifact
Intention is not a forgotten conversation. It's a versioned, structured document that answers "why" and "what" before any "how".
What this changes: Before generating code, intention must be formalized and validated.
2. The contract is executable
A specification that can't be automatically verified isn't a contract, it's a wish.
What this changes: Every requirement has automatable validation criteria (tests, assertions, schemas).
3. No agent validates its own work
An agent that generates code cannot be the source of truth on that code's quality. Validation must always come from an agent that didn't participate in production.
What this changes: Mandatory architecture with 4 minimum roles — Intention, Development, Audit, Retro-Specification.
4. Retro-specification is the system's memory
What was actually implemented must be documented objectively, independently of what the development agent claims to have done.
What this changes: An independent Scribe agent analyzes the code and produces documentation of what actually exists.
5. Foundations precede features
When marginal development cost decreases, investing in architecture becomes rational. We no longer prioritize by "immediate business value" but by "foundation solidity".
What this changes: Sequencing favors what de-risks what comes next, not what impresses stakeholders.
6. Skills are knowledge contracts
AI has potentially outdated training data. Skills inject up-to-date knowledge (documentation, conventions, patterns) and define quality criteria between phases.
What this changes: Each transition (Intention→Dev, Dev→Audit, etc.) is mediated by an explicit skill.
What MADD is not
The 4 MADD Agents
MADD architecture relies on strict separation of responsibilities. Each agent has a unique role and never validates its own work.
Spec Agent — The Intention Architect
Formalizes the "why" and "what". Produces the Intention Document and defines the executable contract with its validation criteria.
Dev Agent — The Developer
Transforms intention into code. Respects the contract, implements features, documents technical choices.
Audit Agent — The Verifier
Audits without complacency. Analyzes produced code, verifies alignment with intention, identifies drifts and risks.
Scribe Agent — The Witness
Documents objective reality. Analyzes what actually exists in the code (not what the dev claims), compares with specs, produces the retro-specification.
Fundamental principle: The Scribe didn't participate in development. Their retro-specification is objective because they have no interest in embellishing or minimizing.
Implementation note: In practice, a 5th role — the Orchestrator — can coordinate the flow between agents, manage dev/audit iteration loops, and determine validation scope. This role can be human or automated.
The MADD Boilerplate provides a ready-to-use implementation with 5 agents, a distributed JSON contract, and an audit recommendation system.
MADD Skills
Skills are knowledge contracts that mediate each transition between agents.
Specification Skill
Guides the Spec Agent in producing structured Intention Documents with verifiable requirements.
Development Skill
Defines conventions, patterns, and constraints that the Dev Agent must respect.
Audit Skill
Lists validation criteria, patterns to detect, audit report format.
Retro-Specification Skill
Guides the Scribe in objective code analysis and production of documentation of what actually exists.
Correction Skill
Frames audit finding corrections with mandatory traceability.
Skills are versioned and evolve with the project. They can be public (framework documentation), project (internal conventions), or delta (temporary knowledge patches).
Getting Started with MADD
Solo (1 person + AI)
- Use 2-3 different models/agents for roles (e.g.: Gemini for specs, Claude for dev, GPT for audit)
- Formalize your intentions before coding
- Never merge without independent audit
- Generate a retro-spec each cycle
Team (2-5 people)
- Assign roles: Intention Architect, Orchestrator, Arbiter
- Centralize skills in a shared repo
- Shared retro-spec as source of truth
Migration from Scrum
- Sprints become intention cycles (variable duration)
- Stories become Intention Docs + Contracts
- Definition of Done becomes the Audit Skill
- Retrospectives become retro-spec consolidations
Contribute
MADD is open source. The methodology evolves through community contributions.
- GitHub: Propose improvements to principles, skills, documentation
- Issues: Report ambiguities, propose clarifications
- Discussions: Share your feedback, use cases
No certification. No paid training. No commercial distortion.
MADD stays open, or doesn't stay.