BMad Method: AI-Driven Development from Ideation to Production

What Is the BMad Method?

BMad (Build More Architect Dreams) is an open-source AI-driven development framework that guides you through the entire software development lifecycle — from a rough idea to working, production-ready code. It does this by orchestrating specialized AI agents through structured workflows that adapt to your project’s complexity.

Unlike using ChatGPT or Claude as a generic assistant, BMad provides:

• Role-specific agents (Architect, PM, Developer) with focused system prompts
• Phase-gated workflows that ensure nothing is skipped
• Artifact generation (PRDs, architecture docs, epics, stories, sprint status)
• An intelligent help system that always knows what comes next

If you have used Claude Code, Cursor, or GitHub Copilot for coding, BMad sits on top of these tools and adds the missing project management and architecture layers.

The Four Phases

BMad structures every project into four sequential phases:

Each phase produces specific artifacts that feed into the next phase. This eliminates the “blank page problem” where developers jump straight into coding without proper planning.

Three Planning Tracks

BMad adapts to project complexity with three tracks:

Quick Flow

Best for: Bug fixes, simple features, clear scope (1-15 stories)

Creates a tech-spec only. The bmad-quick-dev workflow handles planning and implementation in a single pass — perfect for well-understood changes.

BMad Method

Best for: Products, platforms, complex features (10-50+ stories)

Full lifecycle: PRD, Architecture document, UX considerations, Epics and Stories. Each workflow runs in a fresh chat to maintain context quality.

Enterprise

Best for: Compliance-heavy systems, multi-tenant platforms (30+ stories)

Everything in BMad Method plus Security specifications, DevOps planning, and compliance documentation.

The Agent System

BMad uses specialized agents, each with a focused role:

BMad-Help (Your Guide)

The entry point. Inspects your project state, shows available options, recommends next steps. Run it anytime:

bmad-help

Or with context:

bmad-help I have a SaaS idea, I already know all the features. Where do I start?

BMad-Help automatically runs at the end of every workflow to tell you exactly what comes next.

Architect Agent (bmad-agent-architect)

• Creates architecture documents with technical decisions
• Runs implementation readiness checks
• Validates cohesion across planning documents

PM Agent (bmad-agent-pm)

• Creates PRDs from product briefs or direct input
• Generates epics and stories informed by both PRD and architecture
• Maintains decision logs and addenda

Developer Agent (bmad-agent-dev)

• Runs sprint planning (creates sprint-status.yaml)
• Implements stories one by one
• Performs code reviews for quality validation
• Conducts retrospectives after epic completion

Installation

npx bmad-method install

When prompted, select the BMad Method module. This creates:

your-project/
├── _bmad/          # Agents, workflows, tasks, configuration
├── _bmad-output/   # Generated artifacts (PRD, architecture, stories)

For the latest prerelease:

npx bmad-method@next install

The Implementation Cycle

Once planning is complete, implementation follows a repeatable cycle:

1. Sprint Planning — bmad-sprint-planning creates tracking YAML
2. For each story:
   • bmad-create-story — Generate story file from epic
   • bmad-dev-story — Implement the story (fresh chat)
   • bmad-code-review — Quality validation (recommended)
3. After each epic: bmad-retrospective — Learn and adjust

Each step runs in a fresh chat to maintain context quality and avoid prompt pollution from previous work.

Why Fresh Chats Matter

BMad’s insistence on fresh chats per workflow is deliberate. AI coding assistants degrade as context grows — earlier instructions get diluted, hallucinations increase, and the model starts confusing artifacts from different phases. By starting fresh with only the relevant context loaded, each agent performs at peak quality.

Compatibility

BMad works with any AI coding assistant that supports custom system prompts:

• Claude Code (recommended) — Anthropic’s CLI agent
• Cursor — AI-first code editor
• Codex CLI — OpenAI’s terminal coding agent
• Any IDE with AI integration that can load project context

How BMad Compares

When to Use BMad

Good fit:

• Greenfield projects where planning matters
• Teams standardizing on AI-assisted development workflows
• Complex systems needing architecture before implementation
• Anyone tired of “vibe coding” without structure

Overkill for:

• One-off scripts or quick patches (use Quick Flow or raw Claude Code)
• Pure refactoring tasks
• Projects with existing mature architecture docs

Getting Started

1. Install BMad: npx bmad-method install
2. Run bmad-help to see your options
3. Start with Phase 1 (Analysis) or jump to Phase 2 if requirements are clear
4. Follow the guided workflow through all phases
5. Implement story by story with the Developer agent

Resources

• BMad Method Documentation
• GitHub Repository
• BMad Builder — Create custom agents and workflows
• Discord Community
• YouTube Tutorials

Related Content

• Claude Code Bootcamp Workshop
• Claude Code Enterprise Administration
• Karpathy’s CLAUDE.md: AI Coding Discipline
• Context Architecture for AI Agents