Developing for visionOS with GitHub Copilot for Xcode

I have been experimenting with how to apply AI to visionOS development, but when it comes specifically to the Swift development environment, I am starting to settle on the combination of Xcode and GitHub Copilot for Xcode.

First, as long as I am developing for visionOS, I think Xcode is the only real editor choice. I still have expectations for Fleet, but at the moment it is difficult to entrust everything to it. That means there are not many editor-integrated AI tools that naturally come up as options.

GitHub Copilot for Xcode has been easier to use than I expected, and I like that the Chat window does not feel visually out of place alongside Xcode. Another advantage is that it is easy to indicate the target file and related files as context. It is helpful that it understands the structure of the project files.

However, its behavior differs from the GitHub Copilot extension for VSCode. The VSCode version can rewrite code more comprehensively, while the Xcode version only inserts code, which adds a bit of extra work. I hope it gains functionality closer to the VSCode version soon.

As AI-assisted development becomes more common, I have been thinking that the role of engineers may change as well.

In the coming era, people who may be slow at writing code themselves but are extraordinarily fast at reviewing what others have made could be treated as people with a superpower. AI is already writing code in more situations, and the value of being able to review those outputs quickly and give accurate feedback will only increase.

Struggling with conversations with AI has also made me think. For example, ChatGPT sometimes becomes a different kind of difficult from a human. When that happens, instead of stubbornly continuing, it can be faster to discard the chat and start again from the beginning. Perhaps the trick to working well with AI is not expecting too much from it.

As AI development support advances, I suspect more people who say they are not good at managing humans may end up in roles like AI development manager. Even if someone struggles with human relationships, there may be paths where they can build a career by applying development management and optimization skills.

These days there is a lot of comparison information about different AI tools, and it is easy to get lost. Personally, though, I honestly feel that almost any of them is fine. They will all keep competing and evolving, so I do not think today's performance matters too much. It feels more important to pick one that fits you and build your own workflow around it.

So the choice of Xcode + GitHub Copilot for Xcode is simply the best combination for me right now, and another optimal answer may appear someday. For now, I still like it, so I plan to keep digging into it for a while.