Retrospective Guide

Establish effective retrospective mechanisms to continuously improve AI-assisted development practices

Why Retrospectives Matter

AI-assisted development is a continuous learning and optimization process. Retrospectives help you:

• Discover Patterns: Identify scenarios where AI excels and struggles
• Accumulate Experience: Transform individual experience into team assets
• Continuous Improvement: Constantly optimize Prompts, Rules, and workflows
• Risk Control: Promptly discover and fix AI-introduced issues

Retrospective Frequency and Format

Code Review Level (High Frequency)

Frequency: Every PR / Daily

Purpose: Quickly identify problem patterns in AI-generated code

// Can mark with comments
// AI-GENERATED: The following code was generated by Cursor
 
// Or explain in PR description
// 🤖 AI-assisted generated files: src/components/UserCard.tsx

<!-- .cursor/rules/code-style.mdc -->
## Naming Conventions
- API call functions use `fetch` or `get` prefix + resource name
- Examples: fetchUserData, getUserProfile
- Prohibit vague naming like getData, handleClick

Team Level (Regular)

Frequency: Bi-weekly / Monthly

Format: Team retrospective meeting (15-30 minute dedicated topic)

Agenda Template:

## AI-Assisted Development Bi-Weekly Retrospective
 
### 1. Data Review (5 mins)
- Code acceptance rate this period
- Main issue type distribution
- Perceived efficiency improvement
 
### 2. Experience Sharing (10 mins)
- Excellent case showcase
- Pitfall experience sharing
- Newly discovered efficient patterns
 
### 3. Improvement Actions (10 mins)
- Rules update suggestions
- Prompt template additions
- Workflow optimization suggestions
 
### 4. Action Items (5 mins)
- Assign owners
- Set completion dates

Key Retrospective Points

Scenarios Where AI Performs Well

Identify and record scenarios where AI excels:

## AI High-Efficiency Scenarios List
 
### UI Development
- ✅ Static page layouts
- ✅ Form component generation
- ✅ Style adjustments and responsive design
 
### Utility Functions
- ✅ Date formatting
- ✅ Data validation
- ✅ String processing
 
### Documentation
- ✅ API documentation
- ✅ README writing
- ✅ Code comments

Action: Use AI more boldly in these scenarios, reduce manual intervention.

Scenarios Where AI Struggles

Similarly record scenarios requiring caution:

## AI Caution Scenarios List
 
### Database Operations
- ⚠️ Complex JOIN queries
- ⚠️ Transaction handling
- ⚠️ Performance optimization
 
### Business Logic
- ⚠️ Financial calculations
- ⚠️ Permission control
- ⚠️ State machine transitions
 
### Security Related
- ⚠️ Authentication/Authorization
- ⚠️ Data encryption
- ⚠️ Input validation

Action: These scenarios need more detailed Prompts, stricter Review, or manual coding.

Cursor Rules Adjustment Suggestions

Rules optimization points discovered during retrospectives:

Adjustment Checklist Template:

Workflow Optimization Suggestions

Process improvements based on practice feedback:

Case Sharing Mechanisms

Excellent Prompt Case Library

Build a team-shared Prompt template library:

Example: prompts/ui-component.md

Prompt Template:

Based on the following design requirements, generate a React component:
- Component name: [name]
- Functionality: [description]
- UI library: [ui-library]

Requirements:
1. Use TypeScript
2. Component must support className prop
3. Include basic loading and error states
4. Add necessary aria attributes

Effective Workflow Templates

Record validated effective workflows:

Example: workflows/figma-to-code.md

Scenario: Generate frontend code from Figma designs

Steps:

1. Export Figma screenshot (2x resolution recommended)
2. Use Prompt: “Generate React component based on screenshot, use Tailwind CSS”
3. After initial generation, add interaction details
4. Review generated code, focus on:
   • Responsive breakpoints
   • Accessibility attributes
   • Component granularity

Notes:

• Complex animations need separate handling
• Icons should use existing project icon library

Pitfall Records and Solutions

Build a problem knowledge base:

Example: pitfalls/database-query.md

Solution:

1. Explicitly require performance consideration in Prompts
2. Add Rules (Database queries must):
   • Use eager loading to avoid N+1
   • Add index suggestions for complex queries
   • Large data queries must be paginated
3. Focus on database operations during Review

Related Files: .cursor/rules/database.mdc

Continuous Improvement Loop

Establish Tracking Mechanism

## Improvement Tracking Table
 
| ID | Issue | Improvement | Owner | Status | Verification |
|----|-------|-------------|-------|--------|--------------|
| 001 | Deep component nesting | Update Rules | @Zhang | ✅ Done | Nesting reduced 40% |
| 002 | Missing API error handling | Add Prompt template | @Li | 🔄 In Progress | - |
| 003 | Low test coverage | Add test step to workflow | @Wang | 📋 Pending | - |

Technical Debt Radar

Regularly check technical debt in AI-generated code:

# Search for code marked as AI technical debt
grep -r "AI-DEBT" src/
 
# Track quantity trends
echo "This month's AI tech debt: $(grep -r "AI-DEBT" src/ | wc -l)"

Retrospective Checklist

Reference the following checklist for each retrospective:

• Have core metrics changed this period?
• Which excellent cases are worth sharing?
• Which problems appeared repeatedly?
• Do Cursor Rules need updating?
• Do Prompt templates need additions?
• Does the workflow need adjustments?
• Were last retrospective’s action items completed? What was the effect?

Next Steps

Learn how to establish a feedback collection mechanism to systematically collect team feedback data.