I’ve been using GitHub Copilot for VS Code for some time and have found several effective ways to utilize it. Last week, I also got access to Claude and was wondering what the Coding Assistance in Xcode is like. My goal was to build a simple single-page app making one API request and displaying the result. I only had a little time to experiment with this tool, but I would still like to share my impressions of that, as well as my overall experience.
As a developer, I value expanding my knowledge and maintaining full control over my code. I use AI assistance to stay focused and reduce the need to Google small issues. I also use it for tasks like code completion and boilerplate generation. However, since these tools are prediction-based models, their output can be irrelevant or incorrect. Therefore, I consider their responses as a starting point and interpret through my own understanding.
My approach to using these tools efficiently is to create narrowly scoped prompts. This improves accuracy and makes reviewing, verifying, and searching easier when results fall outside my knowledge. I focus on small, controlled requests, mostly involving boilerplate, syntax, or documentation rather than full implementations. I particularly like Copilot’s code completion feature. It’s consistently accurate and speeds up my workflow.
This week, I explored Xcode’s “Coding Assistance” feature. It’s effective at breaking down tasks and implementing them with smooth UI animations. However, it doesn’t perform compilation checks after generating code. Additionally, new prompts can easily override previous ones, which makes sessions feel disconnected. The reason is probably the agent I used, but at this point, I would expect an Apple touch. This is only an initial, limited experience. I’ll continue testing it to see how usable it becomes. For now, though, Copilot in VS Code offers a more reliable experience.