Our Practice Feedback

Real team data and experiences from AI-assisted development practice

Core Metrics Summary

After introducing AI-assisted development, our team (20 developers) collected the following core metrics:

Key Achievements

Significant Coding Efficiency Improvement

The team reported significant efficiency gains in the following scenarios:

• Demo and Prototype Development: Rapid idea validation
• Static Pages: Extremely high UI component generation efficiency
• Utility Functions: Accurate pattern-based code generation
• Standalone Modules: Feature modules with clear boundaries
• Unfamiliar Domains: Such as shell scripts, infrastructure configuration, etc.

Outstanding UI Generation Capability

• High design fidelity
• Reduced time for image slicing and style adjustments
• Good responsive layout foundation generation

Lowered Full-Stack Development Barrier

“Everyone can be a full-stack developer now”

• Easier onboarding for newcomers and cross-domain developers
• Reduced collaboration dependencies, improved individual delivery capability
• Frontend engineers can quickly write backend code and vice versa

Effective Practice Patterns

The team validated the following efficient patterns:

Accelerated Bug Fixes and Refactoring

• Faster problem identification (AI-assisted analysis)
• More comprehensive refactoring suggestions
• Improved test case generation efficiency

Major Challenges

Significant Domain Capability Disparity

AI shows significant performance differences across domains:

┌────────────────────────────────────────────┐
│           AI Capability Distribution        │
├────────────────────────────────────────────┤
│ ████████████████████  Frontend UI / Static │
│ ██████████████████    Simple Business Logic│
│ ████████████          API Development      │
│ ████████              Complex Interactions │
│ ██████                Dynamic/Responsive   │
│ ████                  Database Operations  │
│ ███                   Agent Development    │
└────────────────────────────────────────────┘

Excellent Performance:

• Frontend UI / Static pages
• Simple business logic
• CRUD interfaces

Requires Significant Manual Coding:

• Complex database operations
• Agent development
• Complex interaction logic
• Dynamic effects / Responsive layout details

Maintainability Issues

Common issues:

• Over Design: Unnecessary abstraction layers
• Deep Nesting: Complex component structures
• Inconsistent Naming: Mismatched with existing project style
• Duplicate Code: Similar code across files not abstracted

User Experience Pain Points

Process and Knowledge Management Gaps

• Lack of experience accumulation and unified practices (everyone explores alone)
• No effective sharing and governance mechanism for Prompts/Rules
• Story cards/Tasking difficult for AI to use directly, low structure
• Testing habits degraded (rarely TDD, tests added later)

Future Improvement Directions

Based on feedback, we identified the following priority improvement directions (ranked by importance):

1. Standardize Task Breakdown and Description Methods

Problem: Story cards and Tasking formats are inconsistent, AI has difficulty understanding and executing directly.

Improvements:

• Establish Tasking templates and standards
• Explore Evaluation First / Schema-driven patterns
• Make task descriptions more structured and AI-friendly

2. Establish Cursor Rules Governance Mechanism

Problem: Rules are valuable but lack maintenance mechanism, easily become outdated or bloated.

Improvements:

• Establish Technical Governance process
• Regular review and update of Rules
• Categorized management (general rules vs project-specific rules)

3. Optimize UI Generation Pipeline

Problem: Figma MCP generates code with redundant classes and deep nesting.

Improvements:

• Optimize Figma-to-code conversion rules
• Establish component mapping tables
• Post-processing scripts to clean redundant code

4. Build Team AI Experience Library

Problem: Good practices are scattered in individual experience, cannot scale.

Improvements:

• Build Prompt library (reusable Prompt templates)
• Project knowledge graph
• Persistent storage for technical decisions

5. Tackle Complex Scenarios

Problem: Weaker performance in Agent, database, complex interaction domains.

Improvements:

• Targeted Rules optimization
• Explore further evolution of Draft-Final pattern
• Build domain-specific Prompt templates

6. Improve Code Quality Assurance

Problem: AI-generated code may introduce technical debt.

Improvements:

• Consider LLM-assisted Review Pipeline
• Dual verification for critical code
• Establish “technical debt radar” to track compromised code

One-Line Summary

Recommendations for Your Team

If your team is introducing AI-assisted development:

1. Start with Strong Domains: UI, static pages, CRUD APIs are low-risk, high-reward starting points
2. Establish Feedback Mechanisms: Start collecting data from day one
3. Invest in Cursor Rules: This is the highest ROI investment
4. Manage Expectations: Complex domains require more human intervention
5. Iterate Continuously: AI-assisted development is a continuous optimization process

Next Steps

Learn how to establish a feedback collection mechanism and start collecting your team’s data.