Taeho Kim
Menu
HomeWritingNotesReadingAbout
KOEN

The Current State and Problems of AI Coding

Since AI arrived, many people can easily build websites and apps. But most people skip the formal software development process and simply ask AI tools like Cursor to "build me an app" or "clone this site." This fast approach works for demos, but it doesn't produce production software suitable for real-world use.

"Most people skip the formal software development process and just go to Cursor and ask it to build an app."

These limitations have driven the industry beyond simple vibe coding toward sophisticated context engineering.

A New Paradigm: The BMAD Method

This video introduces a new framework called the BMAD Method. With this approach, AI can divide roles and collaborate as if an entire software development team were working inside an IDE (Integrated Development Environment). Whether you use Cursor AI, Claude Code, or any other tool, this framework lets you build software of any scale.

"Now you can make it feel like an entire software development team is working inside your IDE."

What Is the BMAD Method?

  • BMAD stands for Breakthrough Method for Agile AI-driven Development.
  • It applies the agile development methodology that real software engineers use to AI.
  • A GitHub repository and complete user documentation are provided so anyone can follow along.

"The BMAD Method lets you develop in small units, test, and incrementally deploy -- exactly like a real software team works."

The Core of Agile Development

The BMAD Method follows the standard agile development process:

  1. Write a PRD (Product Requirements Document):
    • List only the needed features without technical details.
  2. Break into small work units:
    • The team splits each feature into small tasks (epics, stories, etc.).
  3. Develop and test per task:
    • Develop and test each task one by one.
  4. Incrementally deploy:
    • Deploy completed parts sequentially.

"Agile development is about breaking software into small pieces, building each one, testing it, and completing it incrementally."

BMAD has AI agents divide these roles and execute them automatically.

How to Actually Use the BMAD Method

1. Preparation

  • Required files:
    • PRD file
    • Architecture file (architecture.md)
  • Download example files from the GitHub repository.
  • Upload files to an AI chatbot (ChatGPT, Claude, Gemini, etc.) and provide operational guidelines.

"The operating system guidelines are attached, so I told the AI to stay in character and follow them exactly."

2. Brainstorming

  • Use the brainstorm command to flesh out app ideas with AI.
  • Example: AI suggests features, UI, widgets, and more for an iOS productivity app.

"It suggested sample interactions for UI design, color-coded categories, smart filters, and tap-to-expand task views."

  • Feature matrices and roadmaps (features organized by version) are also auto-generated.

3. Writing the PRD

  • Switch to the PM (Product Manager) role and use the create doc command to write the PRD step by step.
  • Each step provides decision menus that a real software team would make.
  • Even if you enter something wrong, the AI gives immediate feedback.

"I accidentally pressed 0, and it immediately told me that's not a valid option."

4. Architecture Design

  • Switch to the Architect role to design the app's tech stack and component connections.
  • Complete the full architecture plan in interactive mode.

The Actual Development Process in an IDE

1. Folder and File Structure

  • Manually create a docs folder and place the PRD and architecture files in it.
  • The remaining folders (cla, cursor, etc.) are auto-generated by the agents.

2. Roles by Agent

  • PO (Product Owner) Agent
    • Initialized first.
    • Uses the shard doc command to break PRD and architecture files into small manageable units.

"Shard splits PRD and architecture documents into small indexed pieces so you can follow them step by step."

  • Scrum Master Agent
    • Uses the draft command to review the PRD.
    • Automatically generates epics if they haven't been created yet.

"Epics are large chunks of the PRD. Each epic contains multiple stories."

  • Dev (Developer) Agent
    • Selects a story to implement and starts development.
    • Each story is further broken into sub-tasks so the AI can follow clearly.

"When a story is complete, its status changes to 'review pending' -- it's not 'done' yet. It must pass testing."

  • Testing Agent
    • Uses the review command to automatically check code against requirements.
    • If no issues are found, it changes the story status to "done."

"The testing agent scans the codebase to verify everything required by the story has been implemented."

3. Iterating Like a Real Team

  • Each agent should ideally start in a new chat session (since files are large and context can get tangled).
  • When one story finishes, you move to the next, iterating like a real agile team.

"It progresses exactly like a real software team works -- one step at a time."

Installation and Tips

1. Installation Process

  1. Copy the npx command from the GitHub repository.
  2. Paste and run the command in a terminal (Claude Code, Cursor, etc.).
  3. Enter the project path.
  4. Select BMAD agile core system from the installation options.
  5. Choose whether to split PRD and architecture docs into multiple files (recommended: yes).
  6. Select the IDE(s) to use (multiple selections possible).
  7. Choose whether to include the web bundle (in this example: no).
  8. Once installation is complete, all rule files are generated in the project.

"After installation, type PM and the PMDC file loads immediately, letting you follow the entire BMAD workflow."

2. Additional Tips

  • Using a tool called Git ingest, you can convert a GitHub repository into an LLM-friendly format, making it easy to get full workflow explanations from AI like Claude.

"Paste the git ingest link into Claude and ask any question -- it explains everything step by step."

Conclusion

By leveraging the BMAD Method, AI can divide roles like a real software team and conduct development systematically and iteratively. This approach is useful not just for developers but for anyone who wants to build and deploy software.

"This workflow isn't just for software developers. Anyone who wants to build and deploy software can use it."

Key Concepts

  • AI Coding
  • Agile Development
  • BMAD Method
  • PRD (Product Requirements Document)
  • Architecture Design
  • Agent Role Division (PO, PM, Architect, Dev, Testing, etc.)
  • Automated Iterative Development
  • Same Workflow as a Real Software Team
  • GitHub, Git ingest, npx Installation
  • IDE Integration (Cursor, Claude Code, Windsurf, etc.)

Related writing

Harvest•Engineering Leadership · AI•English

Claude Code Internal Architecture Analysis

A source map leak in Claude Code's npm package exposed Anthropic's internal codebase — revealing 8-layer security, 4-stage message compression, cost-aware error recovery, and unreleased features like voice mode and multi-agent orchestration.

Apr 1, 2026Read more
Harvest•Engineering Leadership · AI•English

Why Agent-Era Skill Standardization Changes Everything

A walkthrough of Skills as org-wide AI infrastructure: four ecosystem shifts, how specialist stacks and orchestrators work, practical authoring rules (description, one-line gotcha, reasoning), agent-first design, three-tier ops, community repos—and why perfection, depth, and stamina still compound.

Mar 31, 2026Read more
Harvest•Engineering Leadership · AI•English

AX Roadmap That Leads to Results: Connecting Individual Efficiency to Organizational Productivity

This webinar by Flex team's CCPO examines why 'using more AI doesn't automatically improve organizational outcomes' through structural analysis. Drawing from real experiments and failures in measurement, sequencing, and organizational adoption, it presents an AX design strategy focused on solving bottlenecks from the last mile - SSOT, evaluation environments, validation, and access control - concluding that changing bottlenecks, verification, decision-making, and collaboration structures matters more than increasing output volume.

Mar 28, 2026Read more