The problem I found with #GitHubCopilot wasn't its ability to write C++. Not all, but most of the code it suggested would have compiled.¹
No, the problem was that it never knew what I was trying to achieve. If I write a line that looks similar to one elsewhere in the same project, I don't want an #LLM to suggest a copy of the line that came next last time I did this. If two lines are the same then they should be factored out into a method or a class. Don't! Repeat! Yourself! And, if no refactoring is warranted then it means that suggesting a repeat of the same thing I did last time is wrong and is a time-wasting, attention-sapping distraction.
So the LLM almost always ended up encouraging me either to duplicate code (bad) or to accept the wrong code (worse).
¹ It occasionally miscounted parentheses, hallucinated methods, or got argument lists wrong.
#GitHubCopilot und Co.: Warum #KI-#Coding-Tools an #komplexen Projekten scheitern https://t3n.de/news/github-copilot-und-co-warum-ki-coding-tools-an-komplexe-projekten-scheitern-1691037/?utm_source=twitter.com&utm_medium=social&utm_campaign=social-buttons via @t3n
5 ways to transform your workflow using GitHub Copilot and MCP.
Introducing the Awesome GitHub Copilot Customizations repo | by Matt Soucoup & Aaron Powell.
https://devblogs.microsoft.com/blog/introducing-awesome-github-copilot-customizations-repo
How To Get Better AI Reponses from GitHub Copilot in Seconds | with James Montemagno.
https://www.youtube.com/watch?app=desktop&v=ZohAaUQBDbs&feature=youtu.be
Better Models, Smarter Defaults: Claude Sonnet 4, GPT-4.1, and More Control in Visual Studio | by Rhea Patel.
From pair to peer programmer: GitHub's vision for agentic workflows in GitHub Copilot.