Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

ChatGPT Now Integrates Directly with Xcode and VS Code on Mac

MacRumors

MacRumors

macrumors bot
Original poster
Apr 12, 2001
67,805
38,410


OpenAI has updated the ChatGPT app for macOS with the ability to directly edit code in popular development environments including Xcode, VS Code, and JetBrains tools.

chatgpt-macos-edit-xcode.jpg

The new feature allows the AI chatbot to make changes to code without requiring users to copy and paste between applications. ChatGPT can now read code from these environments and make edits directly within them.

An optional "auto-apply" mode enables ChatGPT to implement changes without additional clicks. Alexander Embiricos, a member of OpenAI's product team, shared a demonstration of the feature on social media and noted that the capability will come to Windows "soon."

The direct code editing functionality is currently available to ChatGPT Plus, Pro, and Team subscribers who update their macOS app. OpenAI plans to roll out the feature to Enterprise, Education, and free users next week.

This update builds on ChatGPT's "work with apps" capability launched in beta last November, which allowed the AI to read code from development environments but not directly modify it.


The change makes ChatGPT a more direct competitor to specialized AI coding tools like GitHub Copilot and Apple's Swift Assist, both of which offer similar functionality.

Article Link: ChatGPT Now Integrates Directly with Xcode and VS Code on Mac
 
Article Link
https://www.macrumors.com/2025/03/07/chatgpt-integrates-with-xcode-mac/
  • Like
  • Wow
Reactions: SFjohn, Michaelgtrusa and amartinez1660
Please can someone confirm, does it work only with 4.5, not o3 mini high? I will edit this message when I have the answer
UPDATE: runs any model
 
Last edited:
  • Like
Reactions: SFjohn and jamesnajera
In my daily work I often help troubleshooting code that clients have simply copy/pasted code from OpenAI and found out it doesn't work as expected. Always subtile problems, especially for JavaScript and Google Apps but always so small problems (for example using string compare on an array) I'm starting to believe OpenAI does this on purpose.
 
  • Like
  • Haha
Reactions: blob.DK and PlayUltimate
carloslemare said:
I'll never trust OpenAI my code... maybe if the code is to develop things for me, but how can I explain to a client that maybe OpenAI has a copy of the code of their system?
Click to expand...
Easy. Your clients probably used ChatGPT themselves and OpenAI already has many copies of that code.
 
matsan said:
In my daily work I often help troubleshooting code that clients have simply copy/pasted code from OpenAI and found out it doesn't work as expected. Always subtile problems, especially for JavaScript and Google Apps but always so small problems (for example using string compare on an array) I'm starting to believe OpenAI does this on purpose.
Click to expand...
Blame that on your clients who don't test the code.
 
  • Disagree
  • Haha
Reactions: jw2002, ArtOfWarfare and dominiongamma
I have found GitHub copilot to be quite useful, but I parked and stayed there, especially with the convenience of Rider and VS Code integration.

Would anybody with more experience suggest that ChatGPT might be better? Or any other for that matter?
I have seen threads online of social media of never heard before IDEs that seem to do magic… full blown web apps with 3D renderers, physics and all for saying one thing.
Copilot definitely doesn’t do that in my experience.
 
  • Like
Reactions: blob.DK
amartinez1660 said:
Would anybody with more experience suggest that ChatGPT might be better? Or any other for that matter?
Click to expand...
Using JetBrains' "Full Line Code Completion" that runs locally inside IDE projects. Took a couple of days for it to get the language model up to speed, but now it's really good. It seems to be project-aware as well, for example the PHP namespacing is spot-on between projects.
 
  • Like
Reactions: parameter, ArtOfWarfare and amartinez1660
The interface looks like crap compared to using the Copilot extension in VS Code. Apple needs to implement the same side panel as Copilot.

Anyways it’s frustrating as hell when the best language models like ChatGPT and Claude generate stuff you didn’t ask for, gaslight you, lie to you (yes, I fixed all the bugs honest!) and just blatantly disconnect or fail in the middle of a big code change.

They can help a lot but they can also drive you mad.

No, you are not going to generate GTA VII in your IDE. There are no models that can work with millions of lines of code. A 3000 line codebase takes a lot of time for them to update. You can sometimes sit there waiting for 20 minutes for a one line change.
 
  • Love
Reactions: blob.DK
They actually beat Apple to it.
Swift Assist is not even in the latest Beta, yet it was supposed to be shipped in 2024.
 
  • Like
Reactions: parameter
ArPe said:
when the best language models like ChatGPT and Claude generate stuff you didn’t ask for, gaslight you, lie to you (yes, I fixed all the bugs honest!) and just blatantly disconnect or fail in the middle of a big code change.
Click to expand...
Those disconnects are frustrating. Maybe these LLMs are a little too human?
 
I don't have this "Works with Apps" icon. How am I supposed to get it to load? I haven't the latest ChatGPT version, and latest VSCode version...
 
carloslemare said:
I'll never trust OpenAI my code... maybe if the code is to develop things for me, but how can I explain to a client that maybe OpenAI has a copy of the code of their system?
Click to expand...

Fuzzball84 said:
I guess everyone just shrugs their shoulders

This is the big question mark with the implementations of AI. What it gives, it takes back.
Click to expand...

You can opt out of having your data trained on when using an individual account and ChatGPT Team, Enterprise and the API are excluded from training by default.
 
mimeArtist said:
The most common reply I seem to get in ChatGPT is 'You're right...'
Click to expand...
It generally goes:

ChatGPT: here is your code

Me: it doesn't compile here is the message

ChatGPT: you're right try this

Me: it still doesn't compile here is the new message

ChatGPT - you're right try this

After a few iterations we generally get something that more or less works
 
  • Like
Reactions: blob.DK
Ishimura said:
You can opt out of having your data trained on when using an individual account and ChatGPT Team, Enterprise and the API are excluded from training by default.
Click to expand...
You can assure me that none of my data has been used to train the model? Are you sure? Are you completely sure?
 
  • Like
Reactions: parameter
Really don’t want to judge about that stuff or anyone using it. Coding since 3 decades my opinion is: if you want or need to code, then simply learn coding, then practice, practice, practice. These tools look nice, but should be used - in opposite to the impression they would help or support inexperienced users or beginners - by those who know what they do, want and at least - but most important to me - UNDERSTAND what they are proposed from these bots. Used as some sort of brainstorming or tipping in the right direction for specific functions, methods or algorithms one doesnt implement regularly, ok. But you have to understand if the proposed code is efficient, safe, and doing what you want to achive. Copy Paste and try compile some snippets you have no clue about, better dont try start or learning coding that way. Old guy over and out :]
 
  • Like
Reactions: dzhukov, jakey rolling and erthquake
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back