The Dark Side of Using AI for Programming

Key Concepts

• AI tools in programming (ChatGPT, Claude, Amazon Q)
• Learning and Development (L&D) cycle
• Documentation-driven development
• AI-assisted development
• Understanding vs. Task Completion
• Code review implications
• Developer skill degradation
• Trade-offs between speed and learning

Learning and Development Cycle Without AI

• Process:
  1. Consulting documentation (e.g., Discord API docs).
  2. Understanding available APIs, integration types, inputs, and outputs.
  3. Experimentation and iterative improvement based on documentation.
• Benefits:
  • Deep understanding of the system (e.g., Discord APIs).
  • Transferable skills applicable to future projects.
  • Knowledge retention and ability to explain design choices.

Learning and Development Cycle With AI

• Process:
  1. Directly querying the AI tool within the code editor (e.g., "How do I integrate with Discord API to publish a notification?").
  2. Receiving code snippets and methods from the AI.
  3. Implementing the AI-generated code.
• Drawbacks:
  • Lack of understanding of why the code works.
  • Inability to justify API choices during code reviews.
  • Focus on task completion over learning and comprehension.
  • Potential for using suboptimal or incorrect approaches.

The Problem: Task Completion vs. Understanding

• The speaker argues that AI tools can lead to a focus on quickly completing tasks without truly understanding the underlying principles or technologies.
• This is especially problematic for newer developers who may blindly accept AI-generated code without critical evaluation.
• Example: The speaker was able to publish a notification to Discord quickly using AI, but didn't understand the reasoning behind the chosen API.
• Quote: "I was able to get it working pretty fast but but I didn't understand why I didn't understand why it was working."

Real-World Implications and Examples

• The speaker mentions seeing examples on Reddit and other social media platforms of developers submitting code generated by AI that is poorly written or uses inappropriate methods.
• This can lead to code reviews where reviewers question the developer's understanding and the quality of the code.
• Example: Code using non-standard approaches when utility methods already exist.

The Speaker's Solution

• The speaker ultimately reverted to the traditional documentation-driven approach.
• By reading the Discord API documentation, the speaker realized that the AI-generated code used a suboptimal method for their specific use case.
• The speaker then reimplemented the code using the correct method, based on their understanding of the documentation.

Conclusion

• AI tools can be helpful for quickly completing tasks, but they can also hinder learning and understanding.
• Developers, especially newer ones, should be cautious about relying too heavily on AI and should prioritize understanding the underlying principles and technologies.
• The speaker's experience suggests that a balance between AI assistance and traditional documentation-driven development is crucial for maintaining and improving programming skills.
• The key takeaway is that understanding why code works is more important than simply getting it to work.