We work in an empirical world
Empiricism holds that knowledge comes from experience and from decisions made on what you can observe. Scrum names the three moves that make that possible: transparency, inspection, and adaptation. A defined process expects the same inputs to produce the same output. A product does not. The sprint, the review, and the retrospective exist so a team can look at an increment and change what they do next.
An AI can draft a plan that sounds complete. The plan stays a claim until a slice is built, someone looks at the result, and the next slice changes because of what they saw. A delivery method has to live in that world. The model is fast. The evidence still arrives after the work.
BMAD already says what it stands on
The BMad Method is Brian Madison's open-source set of agents and workflows for agile AI-driven development. The method describes those agents and workflows as built on agile practice. The case it makes is that the human stays the expert, the AI facilitates, and the spec is the work. Rigor scales to the change in front of the team. Agile is the base. The agents are how that base gets practiced when a model is in the room.
The loop has four phases: analysis, plan, implement, and verification. A vague idea enters at analysis. A small, clear change can enter at build. Verification reviews the result, ties the checks back to the requirements, and runs a retrospective that feeds the next plan. Learning returns to planning. That is inspection and adaptation, with the artifacts written down so the next session can read them.
The practices it was assembled from
The names on the workflows are recent. The moves are ones a scrum master already runs.
Stories, epics, and acceptance criteria. Planning still cuts intent into epics and stories, and each one carries done-when checks. That contract is the one XP and Scrum already used between the people who want the work and the people who build it. The agent reads the contract. The team still owns it.
Refinement before anyone builds. An epic starts thin and is refined before its stories are planned. A story gets acceptance criteria written ahead of implementation when the work needs that gate. The breakdown is an execution plan, and it is expected to change when earlier work reveals a missing constraint. That is backlog refinement. The definition of ready is the agreement a Scrum team already uses for the same moment.
A done that has to be met. Verification keeps the done-when checks, a code review, and a person who marks the ticket done. The retrospective reads the plans, the diff, and the commits, then judges the epic against its acceptance criteria. It also exercises the changed behavior. A green test suite is one input to that check. An epic that missed its criteria closes unaccepted unless a person overrides the call. The definition of done is that standard, and the retrospective is the inspection of whether the increment met it.
A room that holds the tradeoff. Party mode puts product, architecture, development, and UX voices in one conversation before the team commits. Someone still steers. The Mob Elaboration guide already places that room next to Scrum refinement: several perspectives, one decision, the people still accountable for it.
The system that is already there. On an existing codebase the method takes context from the code in front of the team, then plans forward from that. The next epic looks up whether the last retrospective's action items landed. The evidence is the increment, the history, and the written plan. Those are the inputs empiricism already required.
What you still say in the review
When a team picks up BMAD, the new nouns show up first: agents, skills, party mode, story files. The questions that decide whether the epic went well are older than the method.
Was the story refined before someone built it? Did done mean the acceptance criteria? Did the retrospective change the next plan?
BMAD is useful when those answers are written where the next session can read them, and when a person still makes the call. The team still learns from the increment. The next slice changes because of what they saw. That is the empirical loop the agents were built to carry, and it is the same loop the sprint review was already for.