Beginner

Introduction to BMAD

Discover the Breakthrough Method for Agile AI Development — a methodology designed from the ground up for teams building AI-powered applications.

What is BMAD?

BMAD stands for Breakthrough Method for Agile AI Development. It is an AI-first development methodology and framework for building, deploying, and iterating on AI-powered applications.

BMAD provides structured workflows, prompt engineering patterns, and quality assurance frameworks specifically designed for the unique challenges of AI development projects. It builds on Agile principles but adapts them for the non-deterministic, experimental nature of AI systems.

💡
Good to know: BMAD is not a replacement for Agile or Scrum. It is an extension that addresses the gaps traditional methodologies have when applied to AI-centric projects. Teams already practicing Agile can adopt BMAD incrementally.

Why Traditional Agile Needs Adaptation for AI

Standard Agile methodologies were designed for deterministic software where the same code produces the same output every time. AI development introduces fundamental differences:

Challenge Traditional Software AI Development
Predictability Deterministic — same input, same output Non-deterministic — outputs vary between runs
Testing Unit tests with exact assertions Evaluation metrics, statistical validation
Estimation Relatively predictable with experience High uncertainty — prompt tuning may take days or hours
Definition of Done Feature works as specified "Good enough" is a judgment call — 95% accuracy vs. 99%
Debugging Stack traces, breakpoints Analyzing prompts, evaluating model behavior
Dependencies Libraries, services Model APIs, token limits, rate limits, model updates

The BMAD Lifecycle

BMAD organizes AI development into four distinct phases, each with clear objectives and deliverables:

  1. Blueprint

    Define requirements and map them to AI capabilities. Identify which parts of the system benefit from AI, select potential models, and establish success metrics. This is where you answer: "Can AI solve this problem, and how will we know?"

  2. Model

    Engineer prompts, select models, and prototype AI components. This is the experimentation phase where you test different approaches, compare model outputs, and iterate on prompt designs until you achieve target quality levels.

  3. Actualize

    Implement the AI components into production code. Integrate prompts with application logic, build fallback mechanisms, implement caching and rate limiting, and connect to monitoring systems.

  4. Deploy

    Release to production with monitoring and feedback loops. Track quality metrics, gather user feedback, handle model updates, and iterate based on real-world performance data.

BMAD Lifecycle Overview 
Blueprint → Model → Actualize → Deploy
    |                                          |
    +←←←←←←← Iterate ←←←←←←←←←←←←←←+

Each phase feeds back into the previous ones.
AI projects require more iteration than traditional
software — BMAD embraces this reality.

How BMAD Differs from Standard Agile/Scrum

  • Experimentation sprints: BMAD includes dedicated "spike" sprints for prompt engineering and model evaluation, separate from feature sprints.
  • Quality gates: Each phase has AI-specific quality gates (accuracy thresholds, hallucination rates, latency targets) that must be met before progressing.
  • New roles: BMAD introduces AI-specific roles like Prompt Designer and AI QA Engineer alongside traditional development roles.
  • Evaluation-driven: Where Agile focuses on user stories and acceptance criteria, BMAD adds evaluation datasets and statistical quality metrics.
  • Cost awareness: AI inference costs are a first-class concern in BMAD, tracked and optimized throughout the lifecycle.

When to Use BMAD

BMAD is most valuable when:

✅
Use BMAD when: Your project has significant AI components (LLM-powered features, AI agents, prompt-driven workflows). For projects that only use AI peripherally (e.g., a simple chatbot widget), standard Agile with a few AI-specific practices may suffice.
  • AI is a core part of the product, not just an add-on feature
  • The team is building with LLMs, generative AI, or AI agents
  • Output quality is critical and needs systematic evaluation
  • Prompt engineering is a significant part of the development effort
  • The team needs to manage AI inference costs at scale
📚
Prerequisites: Basic understanding of Agile/Scrum methodology and familiarity with AI concepts (LLMs, prompts, APIs). No specific tools are required to apply BMAD principles.