1. Introduction: Treasures Gained from Numerous Failures

Developing with Genspark can lead to countless failures. Failed Twitter API authentication, 4 hours of lost code, performance degradation due to N+1 problems, system freezes from infinite loops—the list goes on.

However, these failures were not in vain. Rather, failures are the best teachers, and the lessons learned from them form the foundation for future development.

What You'll Gain from This Article

  • 10 concrete lessons learned from real failures
  • Specific ways to apply each lesson immediately
  • Understanding "success patterns" in AI development
  • A mindset to transform failures into growth opportunities

In this article, we will introduce 10 lessons based on real experiences, along with specific anecdotes. We hope this will be helpful to those who are about to start AI development with Genspark, or those who have already started but are facing challenges.

2. Lesson 1: Don't Blindly Trust AI

Background of the Lesson

In early development, we blindly implemented code generated by AI. As a result, code using non-existent API methods and code with security vulnerabilities made their way into the production environment.

AI hallucination is an unavoidable phenomenon. AI can confidently provide incorrect information. Particular caution is needed in the following situations:

  • Information regarding the latest APIs and technical specifications
  • Security-related code
  • Performance-critical processing
  • Complex business logic

How to Apply

  • Check Official Documentation: Always verify API specifications with official documentation
  • Code Review: Understand generated code line by line
  • Incremental Testing: Test in small units and confirm functionality
  • Second Opinion: Verify critical code with another AI tool

3. Lesson 2: Regular Chat Screen Transitions

Background of the Lesson

As a result of working on the same chat screen for a long time, the AI started to forget previous instructions or make contradictory suggestions. Especially after exceeding 100 messages, it noticeably becomes "stupid."

Genspark has limitations in context management. The longer you use the chat screen, the more its quality degrades, making regular transitions essential.

How to Apply

  • Monitor Message Count: Transition to a new screen after 80-100 messages
  • Preparation Before Transition: Save important information to AI Drive
  • Create a Summary: Briefly explain the situation at the beginning of the new screen
  • Prepare References: Have quick references ready

4. Lesson 3: Thorough External Saving

Background of the Lesson

Code created over 4 hours disappeared during a chat screen transition. Relying solely on chat history was a fatal mistake.

Chat history is not persistent storage. It is crucial to always save important code and information externally.

How to Apply

  • Utilize AI Drive: Save code, design documents, and notes to AI Drive
  • Git Management: Manage serious projects on GitHub
  • Regular Backups: Save important files every 30 minutes
  • Multiple Storage Locations: Save to both AI Drive and local storage

5. Lesson 4: Habitual Code Verification

Background of the Lesson

The N+1 problem embedded by AI severely degraded performance in the production environment. A bug unnoticed in the test environment became apparent in real-world operation.

Code generated by AI often has issues from the perspective of performance, security, and maintainability. Merely functional operation is not enough.

How to Apply

  • Static Analysis Tools: Automatic checks with PyLint or ESLint
  • Performance Testing: Test with data close to real-world scale
  • Security Audit: Check for vulnerabilities like SQL injection
  • Code Review: If possible, request a review from a third party

6. Lesson 5: Emphasize Error Messages

Background of the Lesson

In early development, we would skim over error messages and simply pass them off to the AI with "fix this." As a result, the same errors occurred repeatedly, and the root causes were left unaddressed.

Error messages are a treasure trove for problem-solving. By accurately understanding errors and communicating them properly to the AI, you can solve problems efficiently.

How to Apply

  • Full Stack Trace: Provide the entire error message to the AI
  • Error Occurrence Conditions: Describe in detail what operations led to the error
  • Environment Information: Specify OS, version, and dependencies
  • Tried Solutions: Also convey what you have already attempted

7. Lesson 6: Incremental Development

Background of the Lesson

In early development, we tried to implement all features at once. As a result, we couldn't pinpoint where bugs were occurring and spent an enormous amount of time debugging.

Build small, test small, improve small—this cycle is the key to success in AI development.

How to Apply

  • MVP Mindset: Build a minimally functional version first
  • Feature-Unit Implementation: Complete one feature before moving to the next
  • Frequent Testing: Confirm functionality immediately after adding code
  • Version Control: Commit at each stage to allow rollback

8. Lesson 7: Importance of Documentation

Background of the Lesson

I couldn't recall the intention behind code created a week ago, and ended up rewriting it from scratch. This was due to omitting comments and documentation.

In AI development, because the development speed is high, the risk of code becoming incomprehensible when reviewed later increases. Proper documentation is essential.

How to Apply

  • In-Code Comments: Describe the intention behind the implementation (why it was done that way)
  • README Creation: Document project overview and setup procedures
  • Design Documents: Document overall structure and important decisions
  • AI Conversation Log: Save important discussions to AI Drive

9. Lesson 8: Using Multiple Tools Strategically

Background of the Lesson

Trying to solve everything with Genspark alone resulted in wasting time in areas it wasn't good at. By strategically using multiple AI tools, efficiency dramatically improved.

Each AI tool has its strengths and weaknesses. It is important to use Genspark, ChatGPT, and Claude appropriately.

How to Apply

  • Real-time Search: Use Genspark for the latest information
  • Stable Code Generation: Use ChatGPT for standard implementations
  • Logical Thinking: Use Claude for complex designs and analysis
  • Understand Tool Characteristics: Grasp the strengths and weaknesses of each tool

10. Lesson 9: Leverage the Community

Background of the Lesson

I tried to solve a problem on my own and struggled with the same issue for several days. When I asked the community, a solution was found within a few hours.

AI development is a rapidly evolving field. By sharing information with the community, you can learn the latest best practices and solutions.

How to Apply

  • Participate in Tech Communities: Gather information on platforms like Qiita, Zenn, Stack, etc.
  • How to Ask Questions: Clearly state the specific situation and error details
  • Give Back Knowledge: Share your own experiences and solutions
  • Track Latest Trends: Follow updates on AI technology

11. Lesson 10: Don't Fear Failure

Background of the Lesson

In the early days, I felt that failure was "embarrassing." However, I realized that failure is the most efficient learning method and a driving force for growth.

AI development is a continuous process of trial and error. Not trying for fear of failure results in a greater loss in the long run.

How to Apply

  • Experimental Development: Actively try new technologies and methods
  • Record Failures: Document what failed and why
  • Share Learnings: Share lessons learned from failures through blogs and articles
  • Mindset: View failures as "learning opportunities"

Importance of Learning from Failure

It is important to share such failures as valuable information. Failures are not something to be ashamed of, but rather valuable steps towards the next success.

12. Conclusion: To Those Starting AI Development

The 10 lessons introduced so far are all learned from actual failures. Knowing these lessons beforehand would have saved a lot of time and energy.

Summary of 10 Lessons

  1. Don't Blindly Trust AI: Don't neglect verification and understanding
  2. Regular Chat Screen Transitions: Transition after 80-100 messages
  3. Thorough External Saving: Always save important information
  4. Habitual Code Verification: Check not only functionality but also quality
  5. Emphasize Error Messages: Don't miss clues for problem-solving
  6. Incremental Development: Build small, test small
  7. Documentation: Record for your future self and others
  8. Strategic Use of Multiple Tools: Optimize efficiency by using the right tool for the job
  9. Leverage the Community: Share knowledge and get the latest information
  10. Don't Fear Failure: Continue to challenge and learn

AI development is still an evolving field. There are no perfect workflows or methods to absolutely prevent failure. However, by being aware of these lessons, you can reduce the frequency of failures and accelerate your learning from them.

Next Steps

Reference Links