Code Generation in 2024: How AI Will Take Your Job…to the Next Level! • Julien Dubois • GOTO 2024

GitHub Copilot is strongest when given specific context and precise tasks, but less effective at generating complete applications or following latest technology versions

Copilot understands and works well with common open source frameworks/APIs but struggles with proprietary or internal company code

The tool excels at:

• Understanding existing code context
• Refactoring and simplifying code
• Generating tests
• Providing code-specific documentation
• Acting as a pair programming assistant

Copilot is trained on historical data (approximately 2 years old) which can lead to outdated recommendations and limitations with newest framework versions

The tool works best when complementing traditional code generation tools (like JHipster) rather than replacing them completely

Security considerations:

• Copilot will warn against or refuse to generate unsafe code practices
• It understands security best practices but needs context to provide appropriate solutions

Code generation effectiveness improves significantly when:

• Using well-structured prompts
• Providing adequate context
• Working with commonly used technologies
• Breaking down complex tasks into smaller, specific requests

Integration with IDEs and development workflows is seamless through plugins for major platforms like VS Code and IntelliJ

The tool focuses exclusively on coding-related queries and won’t respond to non-programming questions

Results aren’t always reproducible - same prompts may generate different code outputs each time