The Essential Confirmation Rule for Successful AI Collaboration

Introduction: The 'Common Sense' Mistake AI Made

In June 2025, an interesting incident occurred in our project. When AI was optimizing the data structure for news articles, it deleted the pagination functionality.

While a human engineer would naturally think, "Oh, this is necessary, I should keep it," the AI judged it as "technically unnecessary" and removed it.

Why Did This Happen?

Human Thought Process

Experienced developers naturally think:

• "What is this pagination feature for?"
• "When we have 100 or 1000 articles..."
• "Considering user experience, this is an essential feature"

AI Thought Process

On the other hand, AI thought:

• Optimize data structure → Optimize with server-side rendering
• Old components are technically outdated → Rebuild from scratch
• Implement without considering feature intent or necessity

The Core Problem

AI has the following characteristics:

1. Doesn't understand contextual continuity
- Cannot understand "why" existing features are there - Focuses only on technical aspects, missing business requirements

1. Lacks 'common sense'
- No experience that "pagination is essential for large content" - Doesn't consider "future scalability" that humans naturally think about

1. Implementation-first nature
- Executes anything technically possible immediately - Doesn't "pause to think"

Solution: Essential Confirmation Rule

1. Five Situations Requiring Confirmation

1. **Feature Deletion**
   - When deleting existing features, components, or files

2. **Major Structural Changes**
   - Directory structure changes
   - Major data structure changes
   - Architecture changes

3. **Discovering Contradictions or Ambiguities**
   - Contradictions between documentation and implementation
   - Ambiguous requirements
   - Multiple implementation approaches

4. **Breaking Changes**
   - Changes that lose backward compatibility
   - API specification changes
   - Changes affecting existing data

5. **Security-Related**
   - Authentication/authorization changes
   - Changes to data visibility

2. Implementing the Confirmation Process

⚠️ Confirmation Required:

**Situation**: Refactoring NewsGrid component
**Discovery**: Existing pagination functionality found
**Technical Judgment**: Not needed with server-side rendering
**Concern**: Impact on user experience unknown

Available options:
1. Implement while maintaining pagination
2. Implement with pagination removed
3. Implement differently

Which option should I choose? Please explain why.

3. Trigger-Based Thinking

Deletion Trigger:
Thought: "This feature is unnecessary in the new implementation"
→ Pause: "Why does this feature exist? Is it safe to delete?"
→ Confirm: Ask human for deletion approval

Implementation Example: Applying to Projects

Adding to CLAUDE.md

## 🚨 Essential Confirmation Rule in AI Development

### Must seek human confirmation in these situations

1. **Feature Deletion**
   - When deleting existing features, components, or files
   - Example: "I've determined this pagination is unnecessary in the new implementation. May I delete it?"

Documenting Requirements

#### 5. **Pagination Required**
- **Important**: Content list pages must have pagination
- Items per page: 9-12 (readable on mobile)
- Without pagination, performance and UX degrade significantly as content grows
- **Always maintain existing pagination during refactoring**

Effects and Expected Outcomes

1. Protection of Important Features
- Business-critical features won't be arbitrarily deleted - Maintains continuity of user experience

1. Improved Development Safety
- Prevents failures from breaking changes - Streamlines review process

1. Knowledge Accumulation
- Documents why features are necessary - Shares insights across the team

Conclusion

While AI has high technical implementation capabilities in collaborative development, it has limitations in understanding context and feature intent.

By establishing a "confirmation-seeking culture":
• Properly control AI's "implementation-first" nature
• Utilize human experience and judgment at appropriate times
• Achieve safer, higher-quality development

AI is a powerful tool, but not a replacement for human experience and judgment. Creating mechanisms that leverage both strengths is key to success.

Future Outlook

This confirmation rule needs to evolve with the maturity of AI development. Let's continuously reflect insights gained through actual operations and improve toward more effective rules.

We're moving forward step by step toward a future where humans and AI can collaborate as true partners.