Key takeaways
- Build is now the common implementation path; larger work adds planning around repeated implementation sessions.
- Party Mode can simulate personas in one model or use independent subagents, depending on mode and host support.
- Build Auto handles one unattended unit; backlog selection and epic coordination belong to an orchestrator.
- The framework is MIT-licensed, but model usage, workflow overhead, and migration effort remain real costs.
FAQ
What is BMAD Method?
An open-source set of skills, agent personas, and planning and implementation workflows used inside AI coding tools. It can support a single change or larger work organized into specs and stories.
Is BMAD only for large teams?
No. Current documentation provides a single-session Build path and says trivial edits can skip the process. Choose planning depth according to uncertainty and risk, rather than a fixed team size.
What changed after the early v6 workflow?
v6.11 renamed Quick Dev to Build and consolidated skills; v6.12 chooses a light or fuller path after investigating the change. Older commands and customizations need migration review.
What does BMAD cost?
The software is MIT-licensed and free to use. Coding-tool subscriptions, model usage, and any optional services are separate; there is no universal per-feature cost.
Executive Summary
BMAD Method packages planning, implementation, and review workflows for AI coding tools. It includes specialized perspectives and optional modules, while leaving product decisions with the user. The current project positions the same approach for both new products and existing codebases.[1]
The important change since June is workflow simplification. v6.12.0, released September 4, 2026, follows v6.11's move from Quick Dev to a common bmad-build implementation path. The latest release chooses process depth after investigating a change and records review-triage decisions rather than silently discarding findings. These are documented behavior changes, not independently measured improvements in delivery speed.[2]
| Attribute | Value |
|---|---|
| Project | BMAD Method, maintained under bmad-code-org |
| License | MIT |
| Verified release | v6.12.0, September 4, 2026; checked September 15 |
| Category | Agentic skills frameworks |
Repository, release, and license sources establish the attributes above.[1][2][3]
Product Overview
Choose the Amount of Planning
| Work | Documented route |
|---|---|
| Obvious, low-risk edit | Make the edit directly if human review is sufficient |
| One coherent implementation unit | bmad-build investigates and implements the change |
| Epic spanning several sessions | Define intent with bmad-spec, break it into stories, run Build for each |
| Larger coordinated initiative | Add the requirements, architecture, and shared context the work needs |
Planning tools are optional inputs selected for missing information, not a compulsory march through every persona. Current guidance explicitly says a PRD is unnecessary when a shorter statement is sufficient.[4]
Build inspects the repository before settling its plan, can ask about unresolved intent, then implements and reviews the change. The documentation distinguishes a light path from a fuller approval path and asks the user to inspect the result before requesting a push or PR.[5]
Personas and Party Mode
bmad-party-mode supports several execution modes. The default session mode has one model voice all personas; subagent uses separate agents for substantive rounds; agent-team uses a persistent team and is documented as Claude Code-only. Unsupported modes can fall back to a simpler mode. A room of named personas therefore does not necessarily mean independent reasoning or concurrent processes.[6]
Ecosystem and Web Bundles
The repository lists Method, Builder, Creative Intelligence Suite, Test Architect, Loop, and Game Dev Studio as its official ecosystem. They serve different purposes; installing the core method does not imply using every module.[1] Web bundles package selected planning workflows for Gemini Gems and ChatGPT Custom GPTs. Their output can inform later coding work, but this review did not measure savings versus planning in an IDE.[7]
Technical Architecture
The stable installation guide continues to document:
npx bmad-method install
It requires Node.js 20.12 or later. Skills including Build use uv; the installer can finish with a missing-uv warning even though those skills cannot run yet. Git is required for external modules installed from Git.[8]
The current repository README additionally advertises Skills CLI and Claude Code/Codex plugin marketplace routes. Because the entry points are evolving, follow the instructions corresponding to the installed distribution and verify that its help and Build skills load.[1]
For existing projects, bmad-project-context can create a verified block in AGENTS.md. The docs recommend skipping this when maintained project instructions already suffice, and using code inspection rather than loading a large, stale description of the system.[9]
Unattended work has a boundary: bmad-build-auto runs one implementation unit and produces a terminal status. It does not pick the next story or manage the backlog. The optional Loop scheduler or another orchestrator handles dispatch; Build Auto requires subagent support and reports blocked when it is unavailable.[10]
Strengths
- Planning can follow risk. The current model supports both bounded changes and shared specifications across several sessions.[4]
- Existing-code guidance is explicit. The documentation emphasizes discovering current conventions and retaining only useful project context.[9]
- Review is part of the delivery path. Build includes review and correction, with distinct treatment of unrelated findings.[5]
Cautions
- Migration work remains. v6.11 renamed implementation skills and consolidated review/research commands. v6.12 makes deprecated shims opt-in on fresh installs and changes context-loading defaults; custom overrides deserve review.[2]
- Host capabilities matter. Persona discussions may run in a single model, while unattended Build needs actual subagents. Do not infer isolation from the word “agent.”[6][10]
- Process can outweigh the change. The project's own Build guide permits skipping trivial edits. Tests, human judgment, and repository permissions still need to fit the work.[5]
What Developers Say
In a review published April 13, 2026, Itzhak Eretz Kdosha, identified as a senior engineer at Palo Alto Networks, described testing BMAD v6.0.3 during February with Cursor and Claude Sonnet 4.5 on one backend feature involving authentication, security, and infrastructure. He also disclosed involvement in the ideation of another SDD tool.[11]
He credited BMAD's adversarial review and refinement but found the full workflow costly to read and slow to complete: six days and about $200 for that run. The streamlined flow was faster. These are one developer's historical observations with an older version, not a current BMAD price or a prediction for another team. The September workflow changes make a fresh evaluation more useful than repeating those scores.[11]
Pricing & Licensing
Software: MIT-licensed, including commercial use under the license's terms.[3] The repository describes its workflows as free and without a paywall.[1]
Operating costs: Model usage, coding-tool subscriptions, and the time spent reviewing artifacts depend on the host and workload. Web bundles change where planning happens; they do not establish a universal cost reduction.[7]
Competitive Positioning and Fit
Compare BMAD with GitHub Spec Kit and Superpowers in the category report. The decision should turn on whether the team wants BMAD's planning artifacts and adaptable implementation path, rather than its number of personas.
Best fit: Work with uncertain requirements, several implementation sessions, or decisions that need to survive between agents.
Poor fit: Trivial changes where the team can inspect the edit directly, or deployments where the required host tools are unavailable.
Viability Assessment
The September release documents both feature evolution and breaking changes. That establishes continuing maintenance, while making version pinning and upgrade review sensible for repeatable team workflows. It does not establish a support SLA or demonstrate superior code quality.[2]
Bottom Line
BMAD has moved beyond the earlier picture of a fixed, heavyweight agile ceremony. Evaluate the current Build path on a representative change, then add planning and orchestration only where the work requires them. This profile is based on documentation and source review; it does not claim a new hands-on benchmark.
Research by Ry Walker • methodology
Sources
- [1] BMAD Method repository
- [2] BMAD Method release history
- [3] BMAD Method MIT license
- [4] BMAD — Choose a planning path
- [5] BMAD — Build a change
- [6] BMAD — Run multi-agent discussions
- [7] BMad Code — Web bundles
- [8] BMAD — How to install
- [9] BMAD — Start in an existing codebase
- [10] BMAD — Autonomous development loops
- [11] Itzhak Eretz Kdosha — I tested three spec-driven AI tools