← Back to library

Transcript: STOP Using Claude Code OR Codex — analysis

Watch video

Chase AI28m 14sTranscript ✅Added May 8, 3:52 pm GMT+8

Source video ID: VdxUKiF8CWI

Transcript

  • 0:00 — you are hamstringing yourself if you are trying to choose between claude code or codeex. Now claude code has owned the AI discourse for months now and that’s because the gap between claude code and the number two option was extremely large but codeex has quietly closed that gap. GPT 5.5 is an amazing model and is arguably better than Opus 4.7. The usage limits are way more generous with OpenAI’s pro plan than with Anthrox Max plans. And yes, that’s still the case
  • 0:31 — even though they doubled the 5-hour limits. They certainly didn’t double the weekly limits, by the way. And the Codeex desktop app is a legitimately good product. Now, that’s not to say that Codex is better than Claude Code. It’s to say that you now have options. And the best play isn’t to sit here and try to choose which one of these two good options is better. The best play is to use both. And luckily for us, getting the best of both worlds is extremely easy to do. It takes literally seconds to set up the Codeex desktop app with
  • 1:02 — the Claude Code terminal running inside of it. And mastering both tools is also very easy because the ven diagram of Codeex and Claude Code is basically a circle. There’s like 99% overlap. So if you learn how to use one of these, you can very easily learn how to use the other. So today I’m going to tell you what you should be thinking about if you’re a Claude Code user who’s trying to dip your toes into the codeex waters. We’re going to do a quick demo where I show you how to use these two tools in tandem. And we’ll have a deeper discussion about why I think you need to
  • 1:33 — be tool agnostic, why we shouldn’t pigeon hole ourselves into one coding agent into one company’s ecosystem because let’s be honest, you owe these companies zero loyalty. So today we’re going to focus on the Codex desktop app. Now, there is a codeex CLI, but in my experience, I found that to get the best of both worlds, it’s easiest if we use the Codex desktop app with the cloud code in the terminal inside of it because you can have a terminal open inside of this app. And the desktop app
  • 2:03 — honestly has some really nice quality of life stuff that I will show you, things like an inapp browser and that sort of thing. So, to use it, you just have to go to openai.com/codeex and the installer takes like 2 seconds. Now, let’s talk about pricing really quick. By and large, if we compare this to anthropic stuff, you’re getting more bang for your buck across the board. So, it’s hard to do some sort of like onetoone thing because token costs are different. GBT 5.5, if we’re talking about cost per million tokens, it’s about the same or slightly more expensive actually than Opus, but it
  • 2:33 — uses less tokens. And then usage depends on the time of day or sort of. There’s a lot of factors. So it’s not like a one for one examination or comparison we can do but big picture you get more with OpenAI. Now things you need to note there’s GPT 5.5 and there’s GPT 5.5 Pro. GPT 5.5 Pro is only available if you are on the $100 or $200 pro plan. If you’re on the $20 or less plan, you can have 5.5 just sort
  • 3:04 — of straight up. 5.5 just straight up is still good. 5.5 Pro is obviously a little bit of a step above and it’s a model that actually beats out Mythos in some benchmarks, but if you’re someone who is coming from cloud code and you’re like, “All right, do I’m already doing 200 bucks a month in Cloud Code. Like, do I need to do a hundred bucks in pro to get the full power?” I would suggest just start with the 20 bucks a month and get your feet wet and see how you like it. There’s no And if you really like it, you can always upgrade to 100. For me, in my situation, I’m on the $100 Pro as well as the max plans with Anthropic.
  • 3:36 — Once you install Codeex, you’ll open it up and you will see something like this. Now, before we do the quick rundown of what you need to be thinking about, a quick word from today’s sponsor, me. As you know, I recently released a Claude code masterass, which is the quickest way to go from zero to AI dev, especially if you don’t come from a technical background. But I also just released yesterday a codeex masterass alongsighted. And it’s for two types of people. It’s for those who are brand new, never done any sort of coding stuff, but want to get into codeex. And
  • 4:06 — it’s also for those of you who are a little more experienced, you’ve been inside of cloud code and you’re trying to figure out, hey, how do I make the transition? And really, how do I use these two tools in tandem? So, it’s pretty much everything we talk about today times 10. So, if you want to get your hands on that as well as stuff like my Aentic OS system OS system, you can find that inside of Chase AI Plus. There’s a link to that in the pin comment. So this is going to be the very quick 5m minute run through of codeex what you need to be thinking about and some sort of differences in terms of the UI very intuitive honestly so very chat
  • 4:38 — GPT coded right we have the prompt window I can add photos and files I can do plan mode right here it’s just a toggle we have the permissions setup very similar to permissions inside of cloud code where we have like bypass permissions auto that sort of thing I can choose the intelligence aka the effort as well as the model right here I can also very quickly see where I am in terms of what folders I’m operating in. They call them projects. I can work locally or in the cloud. I can do different work trees, that sort of thing. So, pretty easy to navigate. Now,
  • 5:09 — let’s quickly go through the settings tab. So, you have general, the work mode, you’re going to want to be on for coding. This is just going to give you more technical detail. Permissions shows up again. This is just saying, hey, do you even want these as options to be shown to you? The answer is yes, because you’re going to want to sit on full access. And then over here in general, most of this just has to do with the environment setup itself. One thing you might notice is right here, follow-up behavior. Q versus steer. We’ll talk about that more in detail later. Just keep it on Q for now. Appearance is
  • 5:39 — exactly what you would think it would be, but down here you have pets, which sound kind of stupid at first, but honestly pretty useful because they’re basically like a visual hook that lets you know if Codeex is working in the background or if it’s ready for you to do something else. So, it’s just like this little thing, right? It goes anywhere on your computer. It sits on top of whatever program you’re using. So, if I close out Codeex, I can still see my guy. And you’ll see it later when we’re actually doing a task. It will have a little text stream
  • 6:09 — so you can see what it’s working on. And then it kind of just goes flat like this when it’s done working. So, honestly, I love notification stuff like this. Like with clawed code, I just have, you know, an audio hook go off every time it finishes a task because I probably lose more time with a genting from just like not getting back to the task after I tell it to do something and then I tab out or walk away. So, you know, hey, use it or don’t, doesn’t really matter. Then you have configuration. They have some stuff with hooks. Looks like I need to update. Over here is sort of the
  • 6:40 — approval policy and sandbox settings. So, this is similar to permissions except on a global level. And then over here for a work ba workspace dependencies. You will want to have codeex dependencies switched on which it should be by default. Then we have personalization. So you can kind of choose your personality. This is not this is not agents.md aka claw.md. So codeex has its own version of cloud.md. It’s called agents.mmd. And again we’ll talk about that more detail in a little bit. Personalization’s kind of similar but not exactly. It’s more like hey I
  • 7:10 — always want you to call me by this name or something like that. There’s also memory. This is similar to memory inside of cloud code which is also on by default. I turn this stuff off. It’s it’s like, hey, if I tell Codeex, hey, I always go to the gym on Tuesdays. And then Tuesday rolls around and I say something like, I don’t know what to do today, you’ll say, oh yeah, you uh you go to the gym on Tuesdays. I don’t really care for that stuff to be honest, but up to you. Then the rest of the stuff is like MCP servers, git environments, works trees. This kind of will depend like depends how technical
  • 7:41 — you are, how deep you want to get into that. Then there’s stuff like browser use and computer use. So computer use you need to be on Mac and then browser use is exactly what it sounds like. Then we also have archive chats and usage. So not too much you have to play around with here. Mainly going to be in general and then appearance and configuration. Up top we got plugins. So CEX has plugins and skills. Similar to Cloud Code, the line between those two is pretty blurred. So plugins by and large
  • 8:12 — are almost like skill packs or MCPs that come from providers themselves that you can easily install. So something like Superbase installs a Superbase MCP and the requisite skills. So if I open up a chat now and said, “Hey, open up or create a database inside of Superbase for me,” it just does it. So same thing with all of these and it includes stuff like Chrome and spreadsheets and presentations and it’s a one-click install. Then we have skills. Works pretty much the same as Claude Code. If you just opened Codeex, you probably get a popup that’s going to say something
  • 8:42 — like, “Hey, we noticed you have all these skills from another coding agent. Would you like to import them?” So, it will import with one click of the button pretty much everything from Claude Code or something like Open Code. So, it’s able to recognize that on your computer. So, that’s another thing that makes it really easy to switch between these tools. It’s not like, “Oh man, I built like this skill army on clawed code. I can’t leave it.” Well, it’s like, “No, actually, you can. It just automatically throws it in here.” And so to use these, you know, you just click on them, you can uninstall them, you can turn them on
  • 9:13 — or off. So again, pretty intuitive. You can also manage them up here, create a skill very easily, and it has its own skill creator skill as well. There’s also the automations tab, similar to routines in cloud code. They have some default ones in here. We can go out here and automatically create a new automation. You can set it up on the work tree or local. You can put it in a specific project time, all that stuff. You also have the ability to just like you would in the terminal with cloud code, just say, “Hey, let’s create an automation using X, Y, and Z.” And it will automatically put it in there. But
  • 9:44 — very simple, very intuitive to visually click around here. Now, in terms of navigating the file structure and the space on your computer, the way it breaks it down is projects in chats. So, right now, I can go inside a project called Autoflow, which is one I was on earlier today, or I can add a new project, or I can just, hey, I click here and I’m in a new chat. So a new chat, it isn’t really in any specific folder. This is just like being in the chat window inside of like cloud code desktop. Like I’m just talking to it like it’s chatgbt. If I want to work in
  • 10:14 — a specific folder or I want to start a new project, we’re going to go to projects. So to do that, very simple. You can click up here. You can start from scratch and it will create a new folder inside whatever you’ve set as default. I usually just do use an existing folder so I can get a little bit more specific about where I want to go. So in here we’ll do new folder and we’ll just call like YouTube demo codeex and then yep hey do you want to import some settings? Sure. Let’s do that. Supporting some recent settings
  • 10:46 — changes I had in cloud code. And now you can see here I’m inside my YouTube demo codeex. We’re working locally on the main branch. You can also see that over here in projects. So I can say hey what’s up? And then you now see that chat down here. Now, this chat is pretty much the same as having a terminal window open because I can stay in the same project. And if I go up here and I just do start new chat, hi again. I now have two chat windows open, which
  • 11:18 — is virtually the same exact thing as me having two terminals open, right? Same sort of process. Open in the same folder, doing their own thing, but still kind of working on the same project. and can see everything between one another. But it’s very easy to kind of keep track of it inside of this UI. I can also click on any of the chats. I can copy them, fork it into local, fork it into a new work tree, rename them, pin it, whatever I want to do. I can also very easily click on the project, the three dots right here, open it and explore. So, you know, actually navigating the
  • 11:48 — chats and navigating your file system and having a mental model of where everything is sitting on your machine, very easy to do. And frankly, that’s pretty much the Codex desktop app. lot of other cool stuff going on here, right? You can see like the branch details, very easy to do stuff like get actions, but like that was pretty much the bulk of it, right? Everything I just told you, you can use that and you can build whatever you want. Now, we talked a little bit earlier about, hey, you can use the terminal inside of here up here on the top right. Toggle terminal. Boom. Here’s the terminal. It’s inside my YT
  • 12:18 — demo codeex project. And then we can just run claude. Boom. I now have Claude Code and Codeex open in the same project. Now, in terms of bouncing them off one another, couple ways you can do that. One, we can, and what we’ll do right now is we’ll have it create some sort of little web app for us. And I can have it plan it in Codeex, take the plan, copy into Cloud Code, see what it says, copy and paste back and forth, do that sort of thing. Or I can have Codeex build something, have Claude Code actually look at the code because they’re inside the same directory,
  • 12:48 — figure out what it says. I’m sure there’s actually much more sophisticated and simple ways to do it than even this where you can sort of set up something automatically. I haven’t messed with that too much. Point is like the infrastructure is here. Really easy to do. We have the best of both worlds. So, let’s do a simple little demo to kind of put it through the paces. We’ll ask it to create sort of a content slash research ideation type web app. One part needs to be able to like pull
  • 13:18 — information from a bunch of sources and give us like possible ideas. Second part, I want it to like be able to synthesize all the information it grabs and come up with content ideas. And then third part, let’s have it create some sort of littleer at the bottom. Maybe like a mini camb board to like keep our ideas in track. So needs to be able to research, ideulate, and then actually organize all this data. So let’s see what it does. We’ll start with codec. So put it in plan mode. Oh, also in terms of invoking skills and things of that nature, pretty much the
  • 13:48 — same, you can do forward slash and you can like call a certain skill. So if I was like front-end design skill, boom, there we go. Or I can also do at. So I could do at like spreadsheets and so now it’s using the spreadsheet plugin. I can also just use natural language and just like with cloud code, it should pick it up. But doing like slash commands and at commands, that’s sort of how you point at different things. And same thing for pointing at specific files or folders. It also works the same way. And one
  • 14:18 — other thing, context. Something to note, the 5.5 Pro has a 258K context window versus Claude Codes 1 million. My take, not really a bad thing cuz most people have no idea how to manage their own context. they live in context raw L and a 25 AK pretty much makes it impossible to do that for very long. Now, it has autoco compaction when you hit 258K. And autocompaction has its own slew of issues, especially when we begin compacting the same conversation
  • 14:49 — over and over and over. But like I just showed you, doing the equivalent of for/clear is literally just starting a new chat, right? Because I pretty much just opened up a new session. So context is is one little difference. Let’s go ahead and give it a problem and see what it says. So, I want to create a web app that sort of does three things. Um, ideally, you can do it all on the same page. On one hand, I wanted to be able to look at AI news over the last 24 hours across, you
  • 15:21 — know, the major web sources as well as stuff like YouTube or Twitter. Um, and then I wanted to consolidate it into a report. Number two, I wanted to be able to take all that information and come up with potential content ideas for me, like what would a title be, what would the general outline be? It can be just kind of like bullet point format as well as like a hook. And then lastly, I think if it had some sort of like scheduleuler, maybe like a mini camb board where I could then be like, okay, like let’s take that idea you came up with and let’s do it today and next idea we can do tomorrow. Something like that.
  • 15:52 — So, let’s kind of walk through that, plan it. And so, now we’ll go through its plan mode. And the plan mode is basically the exact same as Claude codes. It’s going to think about it. It’s going to ask you a series of questions. I’ve noticed in with 5.5 Pro on extra high, it tends to ask quite a few questions, although maybe it was just the projects I was working on. And in terms of speed, it’s a little I think it’s probably a little bit slower than Opus, although, you know, I don’t have like hard numbers on that. It’s just kind of what the vibes have been. At the same time, if
  • 16:23 — I’m more just doing like a back and forth chat, 5.5 feels a lot snappier than Opus. So, if it’s doing a bunch of tool calls, a little bit slower. It’s just chatting rather fast. So here’s the plan. Codex came up with build a green field single user local web app with Nex.js, TypeScript, and SQL linked. The app will have one main dashboard for three flows. Click the last 24 hours of AI signal. Generate a concise report plus YouTube video ideas and schedule selected ideas on a mini canvan board. So no paid APIs, created RSS feeds, and
  • 16:54 — local generation. So easy way to get cloud code involved is we’re just going to copy this, paste it into cloud code, see if it has any other ideas or any blind spots. So I said Codex came up with this plan for our app. What do you think? What’s missing? So Claude Code came back, said the plan solid, but has some gaps, some soft concerns, as well as some nitpicking. So what I’m going to do, come back into Codeex, paste that in there, and just say, “What do you think of this?”
  • 17:24 — And submit. Now, we could continue to have this back and forth forever and for the sake of time, we’ll stop here. But the idea is we now have a second set of eyes on what the heck it is AI is coming up with as a plan. And I think this is super important, especially if you’re someone who don’t who doesn’t come from a technical background, right? Because the problem is you go to AI, you have an idea, it gives you a plan. If you have no idea what right is supposed to look like, you’re kind of just like sick, dude. Thumbs up. go for it. And it could
  • 17:55 — be missing a whole lot. Now, we try to get around that by just like asking more questions, being more thorough, asking things like, “What am I not thinking about? What would an expert ask?” But what if we do all that and we have Claude Code take a look? And this also implies the inverse, having Claude Code build it and bring it into codec. So, if nothing else, it should kind of just give you that warm and fuzzy feeling inside that like, all right, like this does make sense. Multiple AI experts are telling me it’s a solid plan. And Codex even says, “I
  • 18:25 — agree with the critique’s main diagnosis. The original plan would reliably summarize what happened, but the product you actually described needs what is worth making a video about today. That requires trend signals, ranking, and competitor saturation checks, not just ingestion. So, it is making some changes to the plan because of what Claude Code came back with. And obviously, this sort of dual model approach is something you can apply to any part of your project. And so, here’s the new plan with the updates. And like I said for time, we’re just going to execute it after this first pass. 23 minutes 21 seconds and it says it has
  • 18:55 — implemented the full local AI trend planner. What’s in place? Goes through it. Key files verification pass. It created a readme. So I can click on the readme and you can kind of see this like inapp sort of thing here. So can see what it actually wrote up and then it shows all the different files. So if I click on all the files, it will quickly show sort of like what it created. Obviously, it hasn’t deleted anything since it’s its first pass, but it would show that as well. If I click on any of these files, I can also show it in
  • 19:25 — review. Once it’s inside review, there is a like diff viewer. I can do some git actions here. And I just it’s really easy to see what is actually executed. Again, I love the terminal. You probably love the terminal, but the terminal does have some limitations when it comes to the convenience factor of seeing seeing everything all in one place. So before we even have cloud code take a look, let’s say spin up the dev server for me and open it up in the sidebar browser.
  • 19:55 — I think I’ve genuinely gotten like so bad at typing since I’ve been using AI so much and just like voice dictation over the last year. I I actually have completely lost the ability to like type like a single sentence without errors. So, what it’s going to do is it’s going to spin up the dev server and it’s going to show us the actual web page here in the inapp browser, which is nice. And so, now we can see the web page in the browser. I’ll move over here so it’s a little easier for you to see. So, here’s what it created. We have the AI trend
  • 20:25 — planner. We can run a scan. We can ingest stuff, report ideas, and overall I think actually for the first pass kind of went with like this. I think it’s like a sort of brutalist brutalist type approach. I think it looks pretty good. I mean, I kind of like it. I don’t know. Everything’s AI slop these days, right? But mini cam band. Can I drag these? No, can’t drag those. Would like to be able to do that. Signal feed sources. Okay. On the surface, I don’t know if
  • 20:57 — any actually works, but it looks decent at the beginning. So let’s see what happens if I do run full scan fetching sources. And while it’s doing this, what should we do? Well, we should just have Cloud Code take a look at its work. Hey, can you take a look at what Codeex built on its first pass for our application? Any glaring weaknesses? Anything you would change? Is from my understanding everything
  • 21:27 — should actually be wired up properly and working. But do you see any issues that somehow slipped through the cracks? Well, it’s saying didn’t even work. So, we probably have to figure out something with Lama on that end. But also, now a go. Overall, looks pretty cool. And like obviously we can also go onto this on our local browser. One other thing we can do here, it’s kind of reminiscent of things like claw design is you can like annotate things or leave comments. So I could highlight this, leave a comment and say
  • 21:58 — something like, can we make this italic italicized and that then puts it in a annotation right here. And I can add additional follow-up changes or I could just send that right now. And then you kind of have the ability to annotate anything you want. You can also quickly take a screenshot. I do that and then I could paste it in there. So, makes also sort of front end design reviews and iterations like that a lot a
  • 22:30 — lot easier. And hey, there we now have the uh italics. Oh, and Codex did tell me at the beginning, but I was lazy and didn’t even read it was hey to use the local AI generation run lama pull llama 3.18 or set a llama model to a model you already have. I have a few models on my computer. So, we’ll just say hey can you just find it? Hey, so I’m pretty sure I have uh a couple Olama models already on my machine. Can you take a look at which ones those are and then properly wire them up? Also, let’s check to see if these links actually work. So, it says,
  • 23:01 — you know, AI slop is killing online communities from Y Combinator. So, let’s copy that. Oh, yeah. No, it’s a real thing. Very cool. So, down here, Claude came back with its review. So, it said it found several real bugs. Came back with 20 new bugs. Um, and then it says bottom line wiring is correct. The pipeline flows end to end bugs that will surface fast and then has like some timestamp issues competitor selfwarning and some
  • 23:31 — other stuff it brought up. So, you know, pretty good. I mean, 20 things. I I also wonder if telling cloud code like codeex wrote this if it becomes a little extra adversarial, which I would love. probably can actually like make that a skill considering there is a skill for the uh codeex plugin inside of cloud code which is literally called adversarial review. So codeex went ahead and realized I have GLM 4.7 flash on my machine. It wired it up and ran the trend report again. So you can see that here. So this is an actual legit report
  • 24:02 — based on everything it grabbed. So I can look at these video ideas. So why AI slop is killing online communities. If I hit tomorrow, see if it actually puts it down there. uh doesn’t look like it’s kind of spazzing out. So, let’s have Claude code to fix that. Hey, so when I click on one of the video ideas, for example, the why AI slop is killing online communities. If I click on tomorrow, I kind of just get a loading bar and nothing actually
  • 24:33 — happens. Can we fix that? And also secondly, right now in the camb board, I’m not able to actually move things around on the board once they’re in a specific slot like inbox. I don’t have the ability to move them to today or tomorrow, whatever. And so I think this is how you would sort of do this back and forth. Like you can see here, you can have both of them kind of work on something at the same time. You can have something go to clawed code if you feel like Codeex didn’t hack it. And then I think you can have them sort of like make up for certain weaknesses with one another. I mean, usually I would say Claude Code tends to do a little better
  • 25:04 — on the front-end design side of things and design in general than Codeex, but to be honest, I kind of like how it came out on the first pass. So, you know, I could do a whole demo showing you, hey Claude, go do, you know, work on the front end design. And there’s probably a little bit we could work on here, but I think it does pretty good. And what I really wanted you to see from this demo was just like how easy it is to set this up and the fact that it saw things like Cloud Code saw 20 potential bugs that Codeex didn’t pick up on on its first
  • 25:34 — pass. And I think the sort of compound interest of having them check each other over and over and over again throughout a project kind of pays for itself over time because at the beginning like actually we’re using a bunch of tokens to do this. But in the long run, if you’re able to grab these bugs early, pick out these weak points right away, I think in the aggregate, you actually save tokens. And obviously, we can always open this up in our normal browser as well. And it looks like it was able to fix the actual navigation on
  • 26:06 — the cam board. Now the last thing I wanted to talk about very quickly was sort of the bigger idea of being tool agnostic because I think a lot of people they think that oh like I am tool agnostic like I’ll switch to the best tool tomorrow if there’s something better than Opus or I will switch to the other best tool if there’s something better than GPT 5.5. People think that and in reality they don’t actually do that. In reality, what happens is you get used to one tool, you get ingrained in the habit of using it, and then for a lot of people, you become like weirdly tribal about it where it’s like, no,
  • 26:36 — like I’m an anthropic guy. I’m a Claude Code guy. I hate OpenAI. I hate Sam Alden or the complete reverse, you know, as if like they’re a sports team. You shouldn’t care. You really shouldn’t care. You should be very willing to switch all the time. And it’s really easy to switch if you consistently use all of them and use them in tandem because I think we’re getting to the point and I think it’s only going to get worse where it’s like which model is best isn’t so obvious because hey half of us don’t even buy what they’re
  • 27:06 — selling in terms of the benchmarks and b like they’re all kind of converging to be like really really good and the better they get the more they’re kind of blowing past what the average person is doing. The average project that 99% of people are doing can be done by the models that exist today. So what are we going to do with the models that exist five years from now? So I feel like you’re kind of getting further ahead in the long run if you pit them against one another instead of being like I’m a clawed code guy or I’m a codeex guy. And luckily for us, they haven’t created
  • 27:36 — these like walled gardens where I can interact with both simultaneously. If anything, they’ve made it really easy to interact with them. Codeex creating the cloud code plugin and the ability to like bring over skill files and that sort of thing. So I think we’re kind of in a great age for AI. As much as people kind of like to be doomers about it and say, “Oh, like the prices are increasing.” I think in reality we’re we’re in a great spot and it’s only getting better and you can make it better for yourself if you use all the tools available. So that’s where I’m going to leave you guys for today. Hope you were able to get something out of this. As always, let me know what you
  • 28:06 — thought. Make sure to check out Chase AI Plus if you want to get your hands on my Claude code and codeex masterass.