I am reminded of this when I look at entire forks of vscode just to add a LLM code completion assistant.