Why MD Files Are Essential for AI Work – Prompts Alone Aren’t Enough

When creating slides or coding web pages with AI, working only with prompts eventually leads you astray. You start enthusiastically, but as thoughts get complex, you get lost in the countless ideas AI suggests. This is when you need an MD (Markdown) file-based system.

Beyond API documentation, you need to organize foundational policies accurately and clearly in MD files for consistent production. This article introduces MD file-based AI collaboration methods through a real HR policy system case study.

Problems That Arise Working Only with Prompts

Recently, a colleague organized an HR policy system using AI prompts. At first, they were thrilled. AI quickly generated ideas, proposed structures, and filled in content—productivity felt explosive.

But as thinking progressed, problems emerged. They struggled with organization. As the user’s criteria wavered and became unclear, AI kept suggesting new ideas and materials. The user experienced problems confusing the trees for the forest amid vast information, unable to organize it.

Problems that occur when working only with prompts include:

• Direction constantly changing based on conversation flow
• Loss of consistency as AI doesn’t remember previous decisions
• Difficulty grasping overall structure with scattered information
• Confusion when combining into final deliverable

How to Build Structure with MD Files

My colleague and I tried a different approach. We classified the skeleton and main branches first, then distinguished each direction and organized it into MD files with AI.

Step 1: Classify Ideas with Mind Map

When thoughts are scattered, first classify ideas in a mind map format. This is the stage where you visually grasp the overall structure and decide what MD files are needed.

Step 2: Create MD Files by Topic

Organize each classified topic into independent MD files. For example, an HR policy system could be divided like this:

• recruitment-process.md
• evaluation-system.md
• compensation-policy.md
• training-education.md

Each MD file contains clear policies and standards for its topic. What’s important here is maintaining a consistent organization pattern.

MD File Organization Pattern Example

The evaluation-system.md file from the HR policy system we actually worked on was structured as follows:

markdown

# Evaluation System

## Table of Contents
1. Basic Structure
2. Roles and Permissions
3. Deduction Criteria
4. Operation Process
5. System Screen Design
6. Special Situation Handling

## 1. Basic Structure
- Evaluation cycle: Quarterly
- Evaluation target: All employees
- Evaluation method: 360-degree multi-faceted evaluation

## 2. Roles and Permissions
- Evaluators: Direct supervisor, 2 colleagues, self
- Approvers: Team leader, HR team
- View permissions: Self and upper management

## 3. Deduction Criteria
- 3 tardies: -5 points
- 1 unexcused absence: -10 points
- Project delays: -3 to -15 points depending on situation

## 4. Operation Process
1. Evaluation period announcement (D-7)
2. Evaluation progress (2 weeks)
3. Appeal submission (1 week)
4. Final confirmation and disclosure

## 5. System Screen Design
- Evaluation input screen: 5-point scale, comments required
- Result view screen: Graphs and detailed feedback
- Manager dashboard: Team averages, distribution

## 6. Special Situation Handling
- On leave: Evaluation deferred, included in next evaluation after return
- New employees: Evaluation target after 3 months of employment
- Departing employees: Excluded from evaluation

When each MD file maintains the same structure (table of contents → detailed content) like this, AI can understand and process files consistently when reading and working with them.

Step 3: Collaborate and Enhance with Notion

Transfer each MD file to Notion to share and improve ideas concretely with colleagues. The document’s completeness increases through receiving feedback and enhancing content.

For example, if team members give feedback that “the criteria for project delays are vague” regarding “Deduction Criteria,” reflect this in the MD file to add more specific standards.

Step 4: Organize Integration Direction with MD Files

Once individual MD files are complete, organize the integration direction into another MD file explaining how to combine them. Create a file like integration-plan.md to analyze and quantify the overall structure.

markdown

# HR Policy System Integration Plan

## Connection Structure
- Circular structure: Recruitment → Education → Evaluation → Compensation
- Data linkage needed between each file

## Priority
1. Evaluation system (core)
2. Compensation policy
3. Recruitment process
4. Education and training

## Final Slide Composition
1. Overall overview (1 slide)
2. Detailed policies (4 slides)
3. System screen examples (3 slides)
4. Implementation schedule and expected effects (1 slide)

With this organization, when you ask AI to “create slides based on integration-plan.md,” you can receive much more accurate and consistent deliverables.

Advantages of MD File-Based Approach

Working with AI using this organized system provides several advantages.

Consistent Policies and Direction

Since clear standards are written in MD files, AI doesn’t flow in different directions each time. You can create results much better as intended compared to continuously enhancing through vibe prompts and conversation.

AI Can Also Find Direction

With systematic MD files, AI can also maintain consistent direction. While AI gets confused when users are confused, with clear documentation AI works according to those standards.

Clear and Impressive Final Deliverables

In my colleague’s case, after reorganizing work this way, they achieved clearer and more impressive results creating slide reports with AI. This was thanks to systematically organized structure rather than scattered information.

Practical Tips for MD File-Based AI Collaboration

Stage	Task	Tools
Idea Organization	Visualize overall structure with mind map	MindMeister, Xmind
Document Creation	Create MD files by topic	VS Code, Obsidian
Collaboration & Enhancement	Exchange feedback with colleagues	Notion, GitHub
AI Work	Input prompts based on MD files	Claude, ChatGPT

There are also points to watch when working with MD files:

• Dividing too finely actually creates complexity (classify in appropriate units)
• Clearly indicate connection structure between MD files (use links)
• Regularly update integration document to maintain big picture

Prompt vs MD Files – What’s the Difference

The differences between prompt-based work and MD file-based work are summarized as follows:

Category	Prompts Only	MD File-Based
Consistency	Direction changes with each conversation	Consistency maintained by document standards
Structure Understanding	Difficult	Overall structure clear
Collaboration	Limited conversation content sharing	Easy collaboration through Notion, etc.
Final Result	Scattered information	Systematic and high completion

If you simply need quick ideas, prompts alone are sufficient. However, if you need communication with consistent policies and direction, establishing a system with MD files first is much more efficient.

Conclusion

Working only with AI and prompts is exciting at first, but as thinking gets complex, you easily lose direction. Creating a skeleton with MD files first, clearly organizing each topic, then collaborating with AI yields much more consistent and high-completion results.

Like my colleague’s HR policy system case, systematic document management prevents both AI and users from losing their way. Before directly inputting prompts in your next AI work, try organizing direction with MD files first.