Why I Switched From Claude to ChatGPT (without losing anything)

Sabrina Ramonov 馃崉路 13 min路 2,728 words路 12 min read路 English 路Watch on YouTube

This is the full transcript of Why I Switched From Claude to ChatGPT (without losing anything), published on YouTube by Sabrina Ramonov 馃崉. Every paragraph carries the moment it was spoken, so you can click any line to jump straight to that point in the video, search the whole thing for a word, or copy it out.

0:00In this video, I'll talk about six reasons why Chat GPT Codex is superior than Claude Code. Starting with number one, Codex by far has a superior in-app browser. What I mean by that is within the Chat GPT desktop, within Codex, you have a browser built in. So, you can have your chat side by side with the browser window. So, this is Codex open right now. This is an existing chat where I wrote the newsletter, and the prompt I'm running here is I worked on my newsletter draft within Codex, made edits within Codex, iterated on it within Codex. But, now I'm using the built-in browser over here on the right

0:35side to copy my draft over to Substack. The reason why I have to do it this way is because Substack doesn't have a nice API where you can automate the creation of these newsletters. And so, the in-app browser is really awesome for any automation where you have to interact with a website that does not have an API, but still requires you to be logged in. So, here I have to log in to Substack, and it doesn't have an API, so it's a pain. It's also a pain to do things like copy over images, copy over

1:00videos, format it the exact way I want. And it's super easy to try this, by the way. So, just click this plus button, and then click browser, and it'll open a new browser tab. Now, it pasted a bunch of stuff in the correct format for Substack. I'm not going to touch the website yet, cuz it's still working. So, think of any website, basically, where you have to be logged in, okay, number one. And number two, it's just really

1:22annoying because it doesn't have an API. So, what it's doing right now, it's actually going into the settings for this post and updating the SEO options, as well. So, not only is it pasting in the entire article ready to go, it's also optimizing the SEO description and SEO title. And I I'm literally not doing anything. It's all logging into Substack and doing this for me. So, it happened kind of quickly, but it actually added a footer template from Substack. So,

1:47there's a button here called templates. So, it actually clicked that, added my footer, which is formatted like this. So, it can do all of these cool things in the browser. And the best part is you can automate this, which I'll get to in a bit. So, in 3 minutes, it pasted 1,000 words, it copied over all of the images in the correct sections, it had different styling from the prompts, it updated the SEO title and description, it inserted my template footer. So, that's the headline. Here's a screenshot that Codex took. By the way, I didn't take the screenshot. This is what I mean by styled. So, in Substack, you can have

2:19like different stylings for different sentences, and it automatically did that. It has like titles and stuff, links to everything, more pictures that it inserted, and then my footer at the end. It literally did all of this in like 3 minutes. And it's again really nice because Substack does not have an API, it doesn't allow this level of automation. By far, ChatGPT's in-app browser is incredibly superior to Claude plus the Chrome extension. There will be a use case I'll talk about where Claude actually beat ChatGPT, but that was like one use case I could find. The next area is use of your computer. And basically, what it means is ChatGPT and Claude can

2:57use the apps on your computer. So, let's say you have like Spotify, Slack, Discord. So, these are desktop apps. And then AI over here can actually control each of these desktop apps. Like, it can literally open Spotify, find the playlist or podcast you want to listen to, and hit play. It can open some other desktop app on your computer, on Mac only, and then open it, do stuff in it, write data to it, edit things in it, etc. So, this is what computer use means. And technically, Claude has had computer use for a long time, but I personally like never found it to work well. So, I never used it. ChatGPT

3:33Codex's computer use is much smoother, it's much faster, and it actually just generally gets things done without failing, without complaining, and without hallucinating. So, to just to set it up, open the sidebar, click plugins. And then this is where you can see all different kinds of plugins. So, the one you want is here titled computer use. Many of you will also want like Google Chrome, Google Drive, Gmail, spreadsheets, like I have all of these enabled as well. The next I'm going to talk about is how to combine all of these things into a single workflow. But this is how to set it up, okay? Just go

4:04to plugins, computer use, and test it out. Like literally say, "Open up blank app on my computer and do something simple in it." And see what ChatGPT does. And see how fast it is, most importantly. So, these two things are powerful on their own. I don't doubt that. But when you combine it with number three, it becomes like the ultimate replacement for almost all of your workflow automations. And number three is scheduled tasks. The reason why Codex is superior with scheduled tasks is because you get the things like the in-app browser, you get the ability to use the apps on your computer, and you can schedule all of these things

4:42together. So, you could have a scheduled task that runs at 10:00 a.m. and opens a browser, automates a bunch of work in the browser, and then inputs it into an app on your computer. The nice thing too is it's unified in a single place. I never feel like I have to go somewhere else to get these things done. Claud's is kind of fragmented. You have scheduled tasks in Claud Co-work, you have Claud Code routines, but then

5:04routines don't have access to a browser. It was just like annoying how fragmented scheduling was within Claud ecosystem. So, here's a real task that I have running every day. Basically, I'm looking for a really good editor. This is the scheduled task. It's going to go open this website, which I'll show you in a sec. And then it has a bunch of criteria for reviewing candidates. So, here's the website that it opens. The website is functional, it is a marketplace to hire editors, but it's like pretty bare-bones as a website. It does not have an API. I do have to be authenticated to log in. So, now this

5:36scheduled task basically runs every single day and processes new people that I haven't reviewed yet, and it tells me, "Okay, you should talk to these people cuz they seem super legit." This runs every single day, but you can also run it off schedule. So, you can say, [music] run it now and process all 12 and undecided. Okay, so now it's going to use the in-app browser control skill because it requires being signed in and then it's going to record where it starts and it's going to review each person at a time based on my strict hiring criteria and if the person looks good, it's going to move them to the

6:04[music] next step. And this is all scheduled to run with the browser and everything. I'm just running it off schedule right now just to show you. So, just to summarize this schedule task and the components of it. First one is like run daily at 10:00 a.m. Then it said, go to this website which requires me to be logged in. So, it's going to use that in-app browser. Then it's going to review each candidate and then it's going to click a button. So, it's either going to click essentially yes or no depending on whether the candidate fills a bunch of my criteria. And again, the cool thing about this is it's automation

6:36with that in-app built-in browser and I can go back to that same conversation to see what happened because everything's all in one place in this sidebar. You don't have to go someplace else. So, for example, if you use Claude code and you have all these routines set up, like yeah, you can manage it, but you have to go separately to it's like

6:53claude.ai/code/routine. It's just a separate interface. Whereas here in Codex, all of your chats are integrated here. These workflows are daily automations that are running. Like I can just hop into one and see what previously happened in that automation. I can continue the chat. So, combined, these three, in my opinion, are the most powerful reasons why I've from Claude to Codex. Like Codex has truly become my workflow automation hub because of the combination of these three things in a unified [music] interface. The number four reason why Codex is superior to ChatGPT is its built-in visual generation which obviously uses the latest ChatGPT image model. Claude obviously can read images, but it's just

7:36not very smooth when it comes to creating images within it. I just asked it to make like a photo of cat eating a hot dog. Everything's in the same chat. We're used to it already from using ChatGPT web, but it's like so clunky for some reason to try to do the same thing in Claude. So, if you want to try this out, like open any chat in ChatGPT and just say, "Make a 16:9 image for my

7:58business offer in these two colors." Okay, really simple. Fifth reason why I think Codex is superior to Claude is it has a really nice visualization for GitHub. And for those of you who are not familiar with this is, it's basically like a fancy way to save your work. Especially because if you have AI agents touching your files all the time, something may go wrong. So, GitHub allows you to save your work, create a new branch when you're trying something new, so you don't affect the correct stuff in the original branch. You can experiment whatever you want in the new branch, and if it doesn't work out, you

8:31just go back to the original branch. It's kind of like the Avengers. So, here, normally this will display when you're just like normally using it, but if it's not displaying, there's like little button you can click here, which will show a bunch of metadata about your session. The top section is uh GitHub. You know, if you're non-technical, this basically looks scary, but the first line changes just describes like how many changes you've made since you last saved. Main is like the branch you're on, the one with everything working, and then our experimental branch is like where we're trying stuff out, okay? So, we're currently in the main one. And

9:03really all you have to do when you're non-technical is push this button, commit or push. So, I already did it recently, that's why it's grayed out, but every time you are making a bunch of changes, just say commit or push, and that's basically your save button, okay? So, if you're non-technical. So, this is I'd like this a lot, I think it's really nice. I think it's beginner-friendly-ish, like some of it kind of looks scary, but once you get

9:24used to it, it's really, really easy. Again, all you have to do is, once you make a bunch of changes, you're happy with the changes, just click this commit or push button. That's it, and it'll save your work, save everything over to GitHub. So, number six, this is one of my favorite features to use, but if you're really sensitive about token consumption, you may not want to use it, but it's called fast mode. And basically, what it does is instead of running at the normal 1x speed, ChatGPT works even faster. For me as a business owner, time is the most valuable thing, and I am willing to pay a premium for

9:55it. So, I currently am on the $200 per month ChatGPT plan, just like I'm on the $200 per month Claude plan. Fast mode will basically burn more tokens faster because you're getting chat to work faster, but I find it's worth it cuz I hate sitting around waiting. It's super easy to use it. All you have to do is click the model selector here, and then there's a speed. Right now, you can see it's set on fast. Normally, it's set on standard. So, if you want it faster, it says 1.5x speed, more usage. This doesn't mean 1.5x token usage, by the way. This means 1.5x speed, but the

10:29token usage can sometimes be 2x more, 3x more, etc. So, just keep that in mind. Again, I personally use it, especially if it's during the work day when I want things done quickly. If I'm leaving a goal running overnight, then like I probably don't need it. Uh now, I'm going to talk about one use case where Claude actually won, no matter how hard I tried [laughter] to make ChatGPT win. People, I think don't understand I really prefer simplicity. Like if I could have everything in one place, then I would prefer that because I hate context switch. If I could have made chat do this, then I would have, but I could not

11:03figure out how. So, the use case where Claude actually won was in managing my LinkedIn Recruiter. Basically, I told AI, "Go find me senior engineers who have a certain background in competitive programming." And at first, chat was willing to comply. Like it did, probably found me probably 300 candidates, but at some point, chat just refused to work. It was literally like, "I'm not allowed to help you with anything related to recruiting." Like it just refused to do

11:30it, no matter how I worded it, either. It was like, "This is a hypothetical scenario." Sometimes it would do it if I said, "I'm doing this just for networking purposes, not recruiting purposes." [laughter] But, since it is controlling my browser and logged in, it can like literally see I'm in LinkedIn Recruiter going through list of candidates. Okay, so it was it was hard to continually trick ChatGPT, and it would just like stop and complain. So, what I did instead is I just used Claude Code with the Chrome extension. And this blasted through like everything I asked for. Um so, this was one real use case where Claude Code

12:05really did everything, and ChatGPT actually refused to do the work. Which is ironic, though, because like I think the current sentiment is that Claude has tighter ethical controls than ChatGPT, but in this case it was the complete opposite. It was actually Chat who refusing to do this work, and Claude was like, "Just let me go for it." And it it ran for hours and hours and hours continuously without any objection. All right, so if you're interested in importing your work from Claude to ChatGPT, just head over to settings and then click import in the left sidebar, and then select your other AI app,

12:36whether it's Claude Code or Claude Cork. Now, this did not work for me 100%. So, one thing you want to check is to make sure all your skills get transferred. Just tell CodeX like, "Hey, double-check that all my Claude skills got transferred to ChatGPT I think is like {dot} agent or {dot} agents." You want to have CodeX check that it's grabbed all of the {dot} skills and converted them into skills that CodeX can use. So, just to recap, I've migrated almost my entire workflow to CodeX at this point after an extensive [music] 10 days tier testing. Everything from automations to generating visuals to having it work

13:14faster. For me personally, it's the combination of one two and three that makes CodeX way more powerful, easier to use than Claude.

Where these words come from. This is the caption track YouTube holds for this video, written automatically by YouTube rather than by the creator. We read it, tidied the line breaks and laid it out so it can be read. The plain text version is at https://viewrankai.com/tools/youtube-transcript/s9OC9_FWQ-4.txt.

All rights in this video belong to Sabrina Ramonov 馃崉. Watch it on YouTube. If this is your video and you would rather this page did not exist, tell us and we will remove it.