Core Principles of BMAD

The five foundational principles that guide every BMAD project, the BMAD manifesto, and the specialized roles that make AI teams effective.

The Five Core Principles

1. AI-First Design

Design your system architecture with AI capabilities as a primary consideration, not an afterthought. This means identifying where AI adds genuine value early in the planning process and building your architecture to support AI-specific requirements like model switching, prompt management, and output validation.

2. Iterative Prompt Engineering

Treat prompts as code that evolves through testing and refinement. Version your prompts, test them against evaluation datasets, and iterate systematically rather than through ad-hoc tweaking. A prompt that works in development may fail at scale - systematic iteration catches these issues early.

3. Continuous Evaluation

Establish quantitative quality metrics and measure them continuously. Unlike traditional software where tests pass or fail, AI outputs exist on a quality spectrum. Define acceptable thresholds, measure regularly, and set up alerts when quality degrades.

4. Human-in-the-Loop

Maintain human oversight at critical decision points. AI systems should augment human capabilities, not replace human judgment on important decisions. Design review workflows, approval gates, and escalation paths that keep humans informed and in control.

5. Fail-Fast Experimentation

Embrace rapid experimentation with quick feedback cycles. AI development involves more uncertainty than traditional software. The faster you can test an idea, measure results, and decide whether to proceed or pivot, the more efficient your development process becomes.

Experiment: Can GPT-4 classify support tickets?
Hypothesis: 90%+ accuracy on 5 categories
Time Box: 2 hours
Test Set: 50 labeled tickets

Results:
  Accuracy: 87% (below threshold)
  Categories 1-3: 95% accurate
  Category 4: 72% (ambiguous labels)
  Category 5: 68% (insufficient examples)

Decision: PIVOT
  - Merge categories 4 and 5
  - Add few-shot examples for edge cases
  - Re-test with refined prompt

The BMAD Manifesto

The BMAD manifesto extends the Agile Manifesto with AI-specific values:

BMAD Role Definitions

BMAD introduces specialized roles alongside traditional development roles:

Principle Interactions

The five principles work together as a system:

AI-First Design
  ↓ informs what to build
Iterative Prompt Engineering
  ↓ produces testable outputs
Continuous Evaluation
  ↓ measures quality
Human-in-the-Loop
  ↓ validates decisions
Fail-Fast Experimentation
  ↓ accelerates learning
  ↑ feeds back into design