BMAD Method¶
At a glance¶
| Field | Value |
|---|---|
| Full Name | Build More Architect Dreams (originally "Breakthrough Method for Agile AI-Driven Development") |
| Category | Multi-Agent Orchestration |
| Source | bmad-code-org/BMAD-METHOD / docs.bmad-method.org |
| Author/Org | bmad-code-org |
| License | Open source (proprietary trademark on name) |
| First Released | 2025 |
| Current Version | v6.12.4 stable (October 14, 2026); a V7 preview alpha, distributed as installable skills, is also under active development |
| Stars / Popularity | ~54,600 stars · 6,100+ forks |
| Supported Tools | Claude Code, Cursor, Windsurf, Copilot, Roo Code; MCP Server; Web Bundles for Gemini Gems and ChatGPT Custom GPTs; a Codex plugin marketplace in the V7 preview alpha |
Compatible coding agents¶
| Agent | Support |
|---|---|
| GitHub Copilot (VS Code) | ✅ Supported (V6) |
| GitHub Copilot Coding Agent (github.com) | ❌ Not supported — interactive workflow requires human-driven sessions |
| Claude Code | ✅ Primary |
| Cursor | ✅ Primary |
| Windsurf | ✅ Primary |
| OpenAI Codex (CLI) | ⚠️ Official V7 preview alpha via a Codex plugin marketplace, alongside older community/indirect ports — not yet the same maturity as the V6.12 stable line |
| Gemini CLI | ⚠️ Community/indirect — official support is for Gemini Gems Web Bundles, not CLI runtime parity |
| Roo Code | ⚠️ Community — via V6 cross-platform agent support |
Overview¶
BMAD is an AI-driven agile development framework with named agents for distinct roles, commands, and outputs. Its workflows cover analysis, planning, architecture, and implementation.
Developers direct the process; the agents support decisions rather than make them on the developer's behalf.
Pros & cons at a glance¶
| Pros | Cons |
|---|---|
| ✅ Broad lifecycle coverage | ❌ Heavy upfront investment in spec generation and planning |
| ✅ Scale-adaptive — adjusts planning depth to project complexity | ❌ Steep learning curve — many agents, workflows, and modules |
| ✅ Cross-platform support (Claude Code, Cursor, Windsurf, Copilot) | ❌ Can generate overwhelming volumes of specification documents |
| ✅ Modules for authoring, testing, game development, and creative work | ❌ Team simulation can feel overly ceremonial for solo developers |
| ✅ 54K+ stars and an active Discord community | ❌ Proprietary trademark on the BMAD name |
| ✅ Free and open source | ❌ Four-phase process front-loads significant work before implementation |
BMAD suits projects that need separate product, architecture, and development roles, with a dedicated test role available through the Test Architect module.
Core concepts¶
Specialized Agent Roles. The core BMad Method module (BMM) ships five named agent skills: Mary (Analyst) handles market research and requirements elicitation; John (PM) drives PRD creation; Winston (Architect) designs the technical approach; Sally (UX Designer) owns interaction design; and Amelia (Dev) implements stories with test-first discipline. QA is no longer part of this base roster — it now lives in the separate BMad Test Architect module, run by its own agent, Murat.
Scale-Adaptive Intelligence. BMAD adjusts planning depth based on project complexity, using a lighter process for a bug fix than for an enterprise system.
Module Ecosystem. BMAD's official modules are BMad Method (the core), BMad Builder (skill, workflow, and agent authoring), BMad Creative Intelligence Suite (innovation and storytelling), BMad Test Architect (test planning and release decisions, formerly branded TEA), BMad Loop (unattended epic execution), and BMad Game Dev Studio. bmad-spec and bmad-ux are planning skills inside BMad Method itself, not separate modules.
Party Mode. Multiple agent personas discuss a problem within one session and context window.
How it works¶
BMAD operates through four sequential phases, each driven by the appropriate specialist agents:
Phase 1: Analysis. The Analyst agent conducts domain, market, and technical research before planning begins.
Phase 2: Planning. John (PM) drives the creation of product briefs, PRDs (Product Requirements Documents), and UX design documents to define user needs and the reasons for building.
Phase 3: Solutioning. Winston (Architect) handles architecture design, epic and story breakdowns, and readiness checks before implementation.
Phase 4: Implementation. Amelia (Dev) handles sprint planning, story development, and iteration. Teams that add the BMad Test Architect module get Murat for quality assurance. Retrospectives feed learnings back into the process.
All phases operate through slash commands in the AI IDE. The v6.12 stable line installs with npx bmad-method install and takes bug fixes only; new feature work has moved to a V7 preview alpha.
V7 uses the Agent Skills format instead of an npm-only installer. npx skills add bmad-code-org/BMAD-METHOD --skill '*' installs the full method as skills. Claude Code and Codex users can instead add the plugin marketplace (bmad-code-org/bmad-plugins) and ask the bmad skill to run bmad setup. V7 is pre-beta; its install path, skill names, and docs are still changing.
BMAD also provides Quick Flow for smaller projects, which compresses the planning sequence.
For unattended work, bmad-loop replaces the retired bmad-automator. This separate project is in early open beta and uses Python, not an LLM, to control story execution. Inside the method, the deprecated bmad-dev-auto forwards to bmad-build-auto for one unattended iteration.
Gemini and ChatGPT use Web Bundles for planning. These do not provide CLI or runtime parity with Claude Code, Cursor, or Copilot.
Strengths¶
- Broad lifecycle coverage. BMAD spans analysis, planning, architecture, implementation, QA, and retrospectives through five named BMM agents (Mary, John, Winston, Sally, Amelia) plus Murat in the separate Test Architect module.
- Scale-adaptive. Adjusts planning depth to project complexity, with lighter paths for smaller changes.
- Specialized modules. Agents and modules cover testing, game development, creative work, specification, UX, and unattended iteration.
- Community resources. Roughly 54.6K stars, 6.1K forks, an active Discord community, and multilingual READMEs.
- Cross-platform support (V6). V6 provides integrations for multiple AI coding tools.
- Free and open source. The full framework is available without paywalls or gated content.
Limitations¶
- Heavy upfront investment. More time is required for spec generation and planning than lighter approaches like GSD. The four-phase process front-loads significant work before implementation begins.
- Specification review overhead. Generated documents can take substantial time to review; for some projects, that effort may approach the time needed to write the code directly.
- Steep learning curve. Users must learn which agents, workflows, and modules apply to their task.
- May feel heavy for solo developers. The team simulation — PM, Architect, UX Designer, and more if you add Test Architect — can feel overly ceremonial for a single developer building a small-to-medium project.
- Trademark restrictions. While the code is open source, the BMAD name carries a proprietary trademark, which may affect derivative works or competing distributions.
Best for¶
- Medium-to-large projects needing structured planning and architecture review.
- Solo developers who want team-like discipline. BMAD's agent roles enforce the kind of cross-functional review (PM challenge, architecture review, QA) that normally requires multiple people.
- Projects with complex domain requirements. The Analysis phase (domain research, market research, technical research) is valuable when the problem space itself is poorly understood.
- Teams adopting agile practices with AI. BMAD maps directly to agile workflows (sprints, stories, retrospectives), making it a natural fit for teams already using agile methodologies.
Not ideal for¶
- Quick bug fixes or small scripts. Even with Scale-Adaptive Intelligence and Quick Flow, BMAD's overhead is hard to justify for trivial tasks.
- Projects where speed-to-first-code matters more than specification quality. If the goal is a working prototype in hours, GSD's lighter approach or direct prompting may be more appropriate.
- Environments requiring formal governance and audit trails. BMAD focuses on development process, not enterprise compliance. HVE's constraint-based governance is designed for regulated environments.
- Teams locked in to a single AI coding tool. While V6 improved cross-platform support, the depth of integration still varies by tool.
Community & ecosystem¶
BMAD had roughly 54.6K stars and 6.1K forks on October 27, 2026. The module ecosystem extends into testing, game development, creative work, specification, UX planning, and unattended iteration through the separate bmad-loop project. The project maintains documentation at docs.bmad-method.org and offers multilingual READMEs. Official multi-platform support and Web Bundles have different scopes; Gemini CLI runtime support should still be treated as unavailable, and Codex support sits in the V7 preview alpha rather than the stable v6.12 line.
Comparison notes¶
vs. GSD: GSD Core uses a five-step project loop with fresh task contexts. BMAD adds specialized roles, modules, and four phases, with more upfront planning and broader lifecycle coverage.
vs. Squad: BMAD simulates team roles within a user's AI session, with QA available through Test Architect. Squad runs separate agents in parallel, with persistent memory, shared decisions, and an autonomous work monitor (Ralph). Choose BMAD for role-based planning or Squad for agent coordination.
vs. Superpowers: BMAD uses specialized roles and structured phases to simulate an agile team. Superpowers structures an individual developer's workflow through mandatory TDD, debugging, and review skills. Choose based on whether you need role-based planning or individual implementation practices.
vs. Ralph: Ralph is a minimal autonomous loop. BMAD's unattended options split in two: bmad-build-auto runs one iteration inside the method's own artifacts and roles, while the separate bmad-loop project runs full stories unattended through a deterministic Python control loop — no LLM decides what happens next, only what gets written. Choose Ralph for minimalism and backend freedom; choose bmad-loop when you want unattended, auditable work inside BMAD's artifacts, and accept that it is still in early open beta.