GitHub Copilot has its quirks

I've been using GitHub Copilot with our production codebase for the last 4 months, and here are some of my thoughts:

The Good:

1. Explains Complex Code: It’s been great at breaking down tricky code snippets or business logic and explaining them properly.

2. Unit Tests: Really good at writing unit tests and quickly generating multiple scenario-based test cases.

3. Code Snippets: It can easily generate useful code snippets for general-purpose use cases.

4. Error Fixes: Copilot is good at explaining errors in code and providing suggestions to fix them.

The Not-So-Good:

1. Context Understanding: It’s hard to explain the context to a GenAI tool, especially when our code is spread across multiple files/repos. It struggles to understand larger projects where changes are required in multiple files.

2. Inaccurate Suggestions: Sometimes it suggests installing npm libraries or using methods from npm packages that don’t exist. This is called Hallucination, where AI-generated code looks convincing but is completely wrong.

3. Complex Code: Occasionally, the code it generates is confusing and complex, making debugging harder. In those moments, I wish I had written the logic myself and let Copilot check for errors or bugs.

Overall, GitHub Copilot has been a useful tool, but it has its quirks. When using large language models, the responsibility always stays with the programmer.