Archon/bmad/bmm/docs/enterprise-agentic-development.md
Luis Erlacher 413c6e59a5
Some checks are pending
Build Images / build-server-docker (push) Waiting to run
Build Images / build-mcp-docker (push) Blocked by required conditions
Build Images / build-agents-docker (push) Blocked by required conditions
Build Images / build-frontend-docker (push) Blocked by required conditions
Build Images / build-server-k8s (push) Blocked by required conditions
Build Images / build-mcp-k8s (push) Blocked by required conditions
Build Images / build-agents-k8s (push) Blocked by required conditions
Build Images / build-frontend-k8s (push) Blocked by required conditions
init
2025-11-04 15:31:52 -03:00

19 KiB
Raw Blame History

Enterprise Agentic Development with BMad Method

The paradigm shift: From team-based story parallelism to individual epic ownership

Reading Time: ~18 minutes


Table of Contents


The Paradigm Shift

Traditional Agile: Team-Based Story Parallelism

  • Epic duration: 4-12 weeks across multiple sprints
  • Story duration: 2-5 days per developer
  • Team size: 5-9 developers working on same epic
  • Parallelization: Multiple devs on stories within single epic
  • Coordination: Constant - daily standups, merge conflicts, integration overhead

Example: Payment Processing Epic

  • Sprint 1-2: Backend API (Dev A)
  • Sprint 1-2: Frontend UI (Dev B)
  • Sprint 2-3: Testing (Dev C)
  • Result: 6-8 weeks, 3 developers, high coordination

Agentic Development: Individual Epic Ownership

  • Epic duration: Hours to days (not weeks)
  • Story duration: 30 min to 4 hours with AI agent
  • Team size: 1 developer + AI agents completes full epics
  • Parallelization: Developers work on separate epics
  • Coordination: Minimal - epic boundaries, async updates

Same Example: Payment Processing Epic

  • Day 1 AM: Backend API stories (1 dev + agent, 3-4 stories)
  • Day 1 PM: Frontend UI stories (same dev + agent, 2-3 stories)
  • Day 2: Testing & deployment (same dev + agent, 2 stories)
  • Result: 1-2 days, 1 developer, minimal coordination

The Core Difference

What changed: AI agents collapse story duration from days to hours, making epic-level ownership practical.

Impact: Single developer with BMad Method can deliver in 1 day what previously required full team and multiple sprints.


The Evolving Role of Product Managers & UX Designers

The Future is Now

Product Managers and UX Designers are undergoing the most significant transformation since the creation of these disciplines. The emergence of AI agents is creating a new breed of technical product leaders who translate vision directly into working code.

From Spec Writers to Code Orchestrators

Traditional PM/UX (Pre-2025):

  • Write PRDs, hand off to engineering
  • Wait weeks/months for implementation
  • Limited validation capabilities
  • Non-technical role, heavy on process

Emerging PM/UX (2025+):

  • Write AI-optimized PRDs that feed agentic pipelines directly
  • Generate working prototypes in 10-15 minutes
  • Review pull requests from AI agents
  • Technical fluency is table stakes, not optional
  • Orchestrate cloud-based AI agent teams

Industry Research (November 2025)

  • 56% of product professionals cite AI/ML as top focus
  • AI agents automating customer discovery, PRD creation, status reporting
  • PRD-to-Code automation enables PMs to build and deploy apps in 10-15 minutes
  • By 2026: Roles converging into "Full-Stack Product Lead" (PM + Design + Engineering)
  • Very high salaries for AI agent PMs who orchestrate autonomous dev systems

Required Skills for Modern PMs/UX

  1. AI Prompt Engineering - Writing PRDs AI agents can execute autonomously
  2. Coding Literacy - Understanding code structure, APIs, data flows (not production coding)
  3. Agentic Workflow Design - Orchestrating multi-agent systems (planning → design → dev)
  4. Technical Architecture - Reasoning frameworks, memory systems, tool integration
  5. Data Literacy - Interpreting model outputs, spotting trends, identifying gaps
  6. Code Review - Evaluating AI-generated PRs for correctness and vision alignment

What Remains Human

AI Can't Replace:

  • Product vision (market dynamics, customer pain, strategic positioning)
  • Empathy (deep user research, emotional intelligence, stakeholder management)
  • Creativity (novel problem-solving, disruptive thinking)
  • Judgment (prioritization decisions, trade-off analysis)
  • Ethics (responsible AI use, privacy, accessibility)

What Changes:

  • PMs/UX spend more time on human elements (AI handles routine execution)
  • Barrier between "thinking" and "building" collapses
  • Product leaders become builder-thinkers, not just spec writers

The Convergence

  • PMs learning to code with GitHub Copilot, Cursor, v0
  • UX designers generating code with UXPin Merge, Figma-to-code tools
  • Developers becoming orchestrators reviewing AI output vs writing from scratch

The Bottom Line: By 2026, successful PMs/UX will fluently operate in both vision and execution. BMad Method provides the structured framework to make this transition.


How BMad Method Enables PM/UX Technical Evolution

BMad Method is specifically designed to position PMs and UX designers for this future.

1. AI-Executable PRD Generation

PM Workflow:

bmad pm *create-prd

BMad produces:

  • Structured, machine-readable requirements
  • Testable acceptance criteria per requirement
  • Clear epic/story decomposition
  • Technical context for AI agents

Why it matters: Traditional PRDs are human-readable prose. BMad PRDs are AI-executable work packages.

PM Value: Write once, automatically translated into agent-ready stories. No engineering bottleneck for translation.

2. Automated Epic/Story Breakdown

PM Workflow:

bmad pm *create-epics-and-stories

BMad produces:

  • Epic files with clear objectives
  • Story files with acceptance criteria, context, technical guidance
  • Priority assignments (P0-P3)
  • Dependency mapping

Why it matters: Stories become work packages for cloud AI agents. Each story is self-contained with full context.

PM Value: No more "story refinement sessions" with engineering. AI agents execute directly from BMad stories.

3. Human-in-the-Loop Architecture

Architect/PM Workflow:

bmad architect *create-architecture

BMad produces:

  • System architecture aligned with PRD
  • Architecture Decision Records (ADRs)
  • Epic-specific technical guidance
  • Integration patterns and standards

Why it matters: PMs can understand and validate technical decisions. Architecture is conversational, not template-driven.

PM Value: Technical fluency built through guided architecture process. PMs learn while creating.

4. Cloud Agentic Pipeline (Emerging Pattern)

Current State (2025):

PM writes BMad PRD
   ↓
create-epics-and-stories generates story queue
   ↓
Stories loaded by human developers + BMad agents
   ↓
Developers create PRs
   ↓
PM/Team reviews PRs
   ↓
Merge and deploy

Near Future (2026):

PM writes BMad PRD
   ↓
create-epics-and-stories generates story queue
   ↓
Stories automatically fed to cloud AI agent pool
   ↓
AI agents implement stories in parallel
   ↓
AI agents create pull requests
   ↓
PM/UX/Senior Devs review PRs
   ↓
Approved PRs auto-merge
   ↓
Continuous deployment to production

Time Savings:

  • Traditional: PM writes spec → 2-4 weeks engineering → review → deploy (6-8 weeks)
  • BMad Agentic: PM writes PRD → AI agents implement → review PRs → deploy (2-5 days)

5. UX Design Integration

UX Designer Workflow:

bmad ux *create-design

BMad produces:

  • Component-based design system
  • Interaction patterns aligned with tech stack
  • Accessibility guidelines
  • Responsive design specifications

Why it matters: Design specs become implementation-ready for AI agents. No "lost in translation" between design and dev.

UX Value: Designs validated through working prototypes, not static mocks. Technical understanding built through BMad workflows.

6. PM Technical Skills Development

BMad teaches PMs technical skills through:

  • Conversational workflows - No pre-requisite knowledge, learn by doing
  • Architecture facilitation - Understand system design through guided questions
  • Story context assembly - See how code patterns inform implementation
  • Code review workflows - Learn to evaluate code quality, patterns, standards

Example: PM runs create-architecture workflow:

  • BMad asks about scale, performance, integrations
  • PM answers business questions
  • BMad explains technical implications
  • PM learns architecture concepts while making decisions

Result: PMs gain working technical knowledge without formal CS education.

7. Organizational Leverage

Traditional Model:

  • 1 PM → supports 5-9 developers → delivers 1-2 features/quarter

BMad Agentic Model:

  • 1 PM → writes BMad PRD → 20-50 AI agents execute stories in parallel → delivers 5-10 features/quarter

Leverage multiplier: 5-10× with same PM headcount.

8. Quality Consistency

BMad ensures:

  • AI agents follow architectural patterns consistently (via story-context)
  • Code standards applied uniformly (via epic-tech-context)
  • PRD traceability throughout implementation (via acceptance criteria)
  • No "telephone game" between PM, design, and dev

PM Value: What gets built matches what was specified, drastically reducing rework.

9. Rapid Prototyping for Validation

PM Workflow (with BMad + Cursor/v0):

  1. Use BMad to generate PRD structure and requirements
  2. Extract key user flow from PRD
  3. Feed to Cursor/v0 with BMad context
  4. Working prototype in 10-15 minutes
  5. Validate with users before committing to full development

Traditional: Months of development to validate idea BMad Agentic: Hours of development to validate idea

10. Career Path Evolution

BMad positions PMs for emerging roles:

  • AI Agent Product Manager - Orchestrate autonomous development systems
  • Full-Stack Product Lead - Oversee product, design, engineering with AI leverage
  • Technical Product Strategist - Bridge business vision and technical execution

Hiring advantage: PMs using BMad demonstrate:

  • Technical fluency (can read architecture, validate tech decisions)
  • AI-native workflows (structured requirements, agentic orchestration)
  • Results (ship 5-10× faster than peers)

Team Collaboration Patterns

Old Pattern: Story Parallelism

Traditional Agile:

Epic: User Dashboard (8 weeks)
├─ Story 1: Backend API (Dev A, Sprint 1-2)
├─ Story 2: Frontend Layout (Dev B, Sprint 1-2)
├─ Story 3: Data Viz (Dev C, Sprint 2-3)
└─ Story 4: Integration Testing (Team, Sprint 3-4)

Challenge: Coordination overhead, merge conflicts, integration issues

New Pattern: Epic Ownership

Agentic Development:

Project: Analytics Platform (2-3 weeks)

Developer A:
└─ Epic 1: User Dashboard (3 days, 12 stories sequentially with AI)

Developer B:
└─ Epic 2: Admin Panel (4 days, 15 stories sequentially with AI)

Developer C:
└─ Epic 3: Reporting Engine (5 days, 18 stories sequentially with AI)

Benefit: Minimal coordination, epic-level ownership, clear boundaries

Work Distribution Strategies

Best for: 2-10 developers

Approach: Each developer owns complete epics, works sequentially through stories

Example:

epics:
  - id: epic-1
    title: Payment Processing
    owner: alice
    stories: 8
    estimate: 2 days

  - id: epic-2
    title: User Dashboard
    owner: bob
    stories: 12
    estimate: 3 days

Benefits: Clear ownership, minimal conflicts, epic cohesion, reduced coordination

Strategy 2: Layer-Based

Best for: Full-stack apps, specialized teams

Example:

Frontend Dev: Epic 1 (Product Catalog UI), Epic 3 (Cart UI)
Backend Dev: Epic 2 (Product API), Epic 4 (Cart Service)

Benefits: Developers in expertise area, true parallel work, clear API contracts

Requirements: Strong architecture phase, clear API contracts upfront

Strategy 3: Feature-Based

Best for: Large teams (10+ developers)

Example:

Team A (2 devs): Payments feature (4 epics)
Team B (2 devs): User Management feature (3 epics)
Team C (2 devs): Analytics feature (3 epics)

Benefits: Feature team autonomy, domain expertise, scalable to large orgs


Enterprise Configuration with Git Submodules

The Challenge

Problem: Teams customize BMad (agents, workflows, configs) but don't want personal tooling in main repo.

Anti-pattern: Adding bmad/ to .gitignore breaks IDE tools, submodule management.

The Solution: Git Submodules

Benefits:

  • BMad exists in project but tracked separately
  • Each developer controls their own BMad version/config
  • Optional team config sharing via submodule repo
  • IDE tools maintain proper context

Setup (New Projects)

1. Create optional team config repo:

git init bmm-config
cd bmm-config
npx bmad-method install
# Customize for team standards
git commit -m "Team BMM config"
git push origin main

2. Add submodule to project:

cd /path/to/your-project
git submodule add https://github.com/your-org/bmm-config.git bmad
git commit -m "Add BMM as submodule"

3. Team members initialize:

git clone https://github.com/your-org/your-project.git
cd your-project
git submodule update --init --recursive
# Make personal customizations in bmad/

Daily Workflow

Work in main project:

cd /path/to/your-project
# BMad available at ./bmad/, load agents normally

Update personal config:

cd bmad
# Make changes, commit locally, don't push unless sharing

Update to latest team config:

cd bmad
git pull origin main

Configuration Strategies

Option 1: Fully Personal - No submodule, each dev installs independently, use .gitignore

Option 2: Team Baseline + Personal - Submodule has team standards, devs add personal customizations locally

Option 3: Full Team Sharing - All configs in submodule, team collaborates on improvements


Best Practices

1. Epic Ownership

  • Do: Assign entire epic to one developer (context → implementation → retro)
  • Don't: Split epics across multiple developers (coordination overhead, context loss)

2. Dependency Management

  • Do: Identify epic dependencies in planning, document API contracts, complete prerequisites first
  • Don't: Start dependent epic before prerequisite ready, change API contracts without coordination

3. Communication Cadence

Traditional: Daily standups essential Agentic: Lighter coordination

Recommended:

  • Daily async updates ("Epic 1, 60% complete, no blockers")
  • Twice-weekly 15min sync
  • Epic completion demos
  • Sprint retro after all epics complete

4. Branch Strategy

feature/epic-1-payment-processing    (Alice)
feature/epic-2-user-dashboard        (Bob)
feature/epic-3-admin-panel           (Carol)

# PR and merge when epic complete

5. Testing Strategy

  • Story-level: Unit tests (DoD requirement, written by agent during dev-story)
  • Epic-level: Integration tests across stories
  • Project-level: E2E tests after multiple epics complete

6. Documentation Updates

  • Real-time: sprint-status.yaml updated by workflows
  • Epic completion: Update architecture docs, API docs, README if changed
  • Sprint completion: Incorporate retrospective insights

7. Metrics (Different from Traditional)

Traditional: Story points per sprint, burndown charts Agentic: Epics per week, stories per day, time to epic completion

Example velocity:

  • Junior dev + AI: 1-2 epics/week (8-15 stories)
  • Mid-level dev + AI: 2-3 epics/week (15-25 stories)
  • Senior dev + AI: 3-5 epics/week (25-40 stories)

Common Scenarios

Scenario 1: Startup (2 Developers)

Project: SaaS MVP (Level 3)

Distribution:

Developer A:
├─ Epic 1: Authentication (3 days)
├─ Epic 3: Payment Integration (2 days)
└─ Epic 5: Admin Dashboard (3 days)

Developer B:
├─ Epic 2: Core Product Features (4 days)
├─ Epic 4: Analytics (3 days)
└─ Epic 6: Notifications (2 days)

Total: ~2 weeks
Traditional estimate: 3-4 months

BMM Setup: Direct installation, both use Claude Code, minimal customization

Scenario 2: Mid-Size Team (8 Developers)

Project: Enterprise Platform (Level 4)

Distribution (Layer-Based):

Backend (2 devs): 6 API epics
Frontend (2 devs): 6 UI epics
Full-stack (2 devs): 4 integration epics
DevOps (1 dev): 3 infrastructure epics
QA (1 dev): 1 E2E testing epic

Total: ~3 weeks
Traditional estimate: 9-12 months

BMM Setup: Git submodule, team config repo, mix of Claude Code/Cursor users

Scenario 3: Large Enterprise (50+ Developers)

Project: Multi-Product Platform

Organization:

  • 5 product teams (8-10 devs each)
  • 1 platform team (10 devs - shared services)
  • 1 infrastructure team (5 devs)

Distribution (Feature-Based):

Product Team A: Payments (10 epics, 2 weeks)
Product Team B: User Mgmt (12 epics, 2 weeks)
Product Team C: Analytics (8 epics, 1.5 weeks)
Product Team D: Admin Tools (10 epics, 2 weeks)
Product Team E: Mobile (15 epics, 3 weeks)

Platform Team: Shared Services (continuous)
Infrastructure Team: DevOps (continuous)

Total: 3-4 months
Traditional estimate: 2-3 years

BMM Setup: Each team has own submodule config, org-wide base config, variety of IDE tools


Summary

Key Transformation

Work Unit Changed:

  • Old: Story = unit of work assignment
  • New: Epic = unit of work assignment

Why: AI agents collapse story duration (days → hours), making epic ownership practical.

Velocity Impact

  • Traditional: Months for epic delivery, heavy coordination
  • Agentic: Days for epic delivery, minimal coordination
  • Result: 10-50× productivity gains

PM/UX Evolution

BMad Method enables:

  • PMs to write AI-executable PRDs
  • UX designers to validate through working prototypes
  • Technical fluency without CS degrees
  • Orchestration of cloud AI agent teams
  • Career evolution to Full-Stack Product Lead

Enterprise Adoption

Git submodules: Best practice for BMM management across teams Team flexibility: Mix of tools (Claude Code, Cursor, Windsurf) with shared BMM foundation Scalable patterns: Epic-based, layer-based, feature-based distribution strategies

The Future (2026)

PMs write BMad PRDs → Stories auto-fed to cloud AI agents → Parallel implementation → Human review of PRs → Continuous deployment

The future isn't AI replacing PMs—it's AI-augmented PMs becoming 10× more powerful.



BMad Method fundamentally changes how PMs work, how teams structure work, and how products get built. Understanding these patterns is essential for enterprise success in the age of AI agents.