BMAD Agents vs. BMAD Workflows

They are two distinct but interconnected concepts in the BMAD Method:

Agents

Agents are personas – specialized AI roles with distinct expertise, personality, and behavioral guidelines. Each agent has a name, a display name, a title, and a set of instructions that shape how it thinks and communicates. Think of them as team members on a virtual product team.

Examples from the catalog:

Agent	Display Name	Title	Specialty
analyst	Mary	Business Analyst	Business analysis, research, brainstorming
pm	John	Product Manager	PRDs, epics & stories, requirements
architect	Winston	Architect	Technical architecture, solution design
ux-designer	Sally	UX Designer	User experience design
dev	Amelia	Developer Agent	Story implementation, code review
sm	Bob	Scrum Master	Sprint planning, story creation, status
tech-writer	Paige	Technical Writer	Documentation, diagrams, explanations

An agent by itself doesn’t do anything specific – it establishes the who and how (expertise, tone, decision-making style).

Workflows

Workflows are structured processes – step-by-step procedures that produce a specific artifact or outcome. They define what gets done, in what order, with what inputs and outputs.

Examples: “Create PRD”, “Create Architecture”, “Sprint Planning”, “Dev Story”, “Code Review”.

Each workflow is backed by a workflow file (e.g., workflow-create-prd.md) that contains the detailed instructions, steps, checkpoints, and output format.

How They Connect

A workflow is executed by an agent. The agent provides the expertise and persona; the workflow provides the structured process. In the catalog, every workflow row has both an agent-name and a workflow-file:

/bmad-bmm-create-prd – Workflow “Create PRD” is executed by agent pm (John, Product Manager)
/bmad-bmm-create-architecture – Workflow “Create Architecture” is executed by agent architect (Winston, Architect)
/bmad-bmm-dev-story – Workflow “Dev Story” is executed by agent dev (Amelia, Developer)

Key Differences at a Glance

	Agent	Workflow
What it is	A persona/role with expertise	A structured process with steps
Defines	Who does the work and how they think	What gets done and in what order
Has	Personality, expertise, behavioral rules	Steps, inputs, outputs, artifacts
Reusable?	Yes – one agent can run many workflows	Yes – but each workflow is tied to an agent
Invocation	Loaded via agent command (e.g., `/tech-writer`)	Invoked via slash command (e.g., `/bmad-bmm-create-prd`) or by asking a loaded agent

Two Invocation Styles

Command-based workflows (have a command field) – You run them directly with a slash command like /bmad-bmm-create-prd. The agent is loaded automatically.
Agent-based workflows (empty command field) – You first load the agent (e.g., /tech-writer), then ask it to perform a task using the workflow’s code (e.g., “EC about [topic]” for Explain Concept). Examples include Write Document (WD), Mermaid Generate (MG), and Explain Concept (EC) which all use the tech-writer agent.

In short: agents are the “who”, workflows are the “what”. Together they form the structured BMAD method for moving a project from idea through planning, architecture, and implementation.