MYFORGE / ENGINEERING NOTES
BMAD in AI development: from planning to verification
Updated: 18 September 2026.
How do we use BMAD, choose tools and preserve what we learn during development?
What is BMAD for?
BMAD is an AI-assisted development method connecting problem definition, documented decisions, implementation and review. Using a tool does not by itself establish delivery quality: verification must fit the task.
How much planning is enough?
The official guide starts with how well the intent is defined. A clear task can become a short specification. Missing evidence, a shared product vision or architecture calls for focused research and planning. The size of the work and the reach of decisions determine the documents needed.
How we divide the work at MyForge
We use Gemini and NotebookLM for deep research, web discovery and connecting sources. Research informs a superprompt and brief. With BMAD and Claude we plan requirements, system boundaries and tasks. Codex primarily handles review: examining changes, gaps and testing questions.
Example: a new customer workflow
First we clarify who initiates the request, which data they provide and who confirms it. The plan defines states and roles. Development proceeds in reviewable increments. Review and tests check whether the request belongs to the correct customer, whether confirmation status is clear and what happens when connectivity fails. This is an illustrative workflow; checks are selected for the actual system.
What carries into the next task?
Our Superintelligent Vault connects project decisions and lessons. Crystallisation extracts short, verifiable lessons from a session. After approval, propagation routes them into the appropriate knowledge note, project or skill. The next task can then draw on retrievable prior experience.
