← Everyday AI Podcast – An AI and ChatGPT Podcast
Everyday AI Podcast – An AI and ChatGPT Podcast
Ep 776: OpenAI’s Codex For Beginners: Why it’s better than Claude, ChatGPT and How to Start Using It
00:57:28
Loading summary
Transcript7 lines
- [00:01]
A
This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.
- [00:19]
B
Meet Firefly AI Assistant now live in Adobe Firefly, the all in one creative AI Studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi step workflows across Photoshop, Premiere Express and more. In one conversational interface, you direct the outcome. The Assistant accelerates execution.
- [00:47]
C
Here's a little secret. I don't use ChatGPT much anymore, and I don't think you should either. And that's not because it's not a good AI Chatbot or AI operating system. It's still one of the best. But the reason I don't really use ChatGPT much anymore is actually because of OpenAI's Codex desktop app. That's OpenAI's developer tool turned knowledge worker agent and the simplest way to frame it is Chat GPT works with you, Codex works for you. It can use your computer without interrupting you. It can control your Chrome browser, and it even has a built in file viewer and a built in browser. Yes, OpenAI has also confirmed that Codex is actually their forthcoming super app. So if you want to stay on the cutting edge of what's available in AI, that's actually happening in code, Codex and not necessarily Chat gbt. But maybe you've been scared off just by the word code in Codex or by its developer, only Origins, if that's you, or if you're still not sure if you're really using Codex correctly or if you want to jump in. But if you're not sure how it all works, then today's show is definitely for you. We're going over Codex for beginners, why it's better than Claude code, Claude Cowork, and even chatgpt and how you can actually start using it today, even if you're a non technical person. All right, I'm excited to jump into it. First, here's the big picture. You're not using the most powerful AI system in the world. That is Codex. It's not ChatGPT, it's not Claude on the web, it's not Claude desktop via Cowork or Claude code. It is Codex and it is not even close. So here's the zoomed out version. OpenAI released Codex for the web about a year ago, but then they released the desktop app version in February and since then it has gotten widespread adoption and I think even though I've been screaming about it on the porch Since February. Right. Like old man Jordan on the porch saying, use Codex, kids. I think finally, the rest of the world is finally starting to wake up now to what I've been saying for three months. You can't ignore Codex anymore. And yes, anthropic did beat OpenAI to the punch when it came to desktop versions of their most popular tools, specifically in Claude Code and Claude Cowork. And they were great in their own right. But I don't really use Claude Code or Claude Cowork much anymore because I don't really find a use case and I look for them all the time. And I think OpenAI has really leaned into Codex. Right, the future super app, so heavily that Codex is now measured better than both ChatGPT and Claude's desktop offering. So on today's show, on Wednesdays, we put AI to work. I have on Wednesdays we're doing a nice little live demo. What could go wrong? All right, so we're going to go over the basics of how Codex works and how any free or paid Chat GPT user can try it today. Yeah, you can even be on a free plan and try this. I'm going to give you 35 fast Codex facts that will get you from 0 to 5.0, hopefully in no time. And I'm going to show you how I'm personally using chat GPT vs Codex. And the secret reason your company should probably be using Codex and not using any Claude desktop app. It's one very simple thing that has huge ramifications that most people are shocked when they find out. All right, let's get into it. Welcome to Everyday AI. My name is Jordan Wilson and if you are new here, well, we do this every day. It's a daily unedited live stream, podcast and free daily newsletter helping everyday business leaders like you and me keep up with what's happening in the world of AI. Because it doesn't stop. I tell you what's important, how to use it to grow your company and your career. So it starts here. But make sure, if you want to be the smartest person in AI, our website is your cheat code, your everyday AI dot com. You can obviously go listen and watch to every single episode we've ever done for free there. If you go click on the Episodes tab. But you can also go sign up for our free daily newsletter where we will be recapping the highlights from today's show as well as all of the other AI news that you need to know. All right. Also, yeah, there's so much here we're actually going to do a two parter. So we're going to do part one today and we're going to do part two probably next Wednesday, as long as there's not some, you know, huge other release, in which case we'll push out part two for another week. But this is two parts because it would take me many hours. But I'm going to try to make this one kind of fast basic, but also show you what's possible. So in part two we're going to cover skills, plugins and automations. We're going to go over the basics of computer use, Chrome control and connecting your data. I'm going to show you what goal is. Right. Yeah, that's big, right? The new goal feature inside Codex and why you should be using it. And then I'm going to show you the right way to set up standard and advanced workflows that run on a schedule in our light years ahead of what you could actually do inside of Claude Cowork or Claude Code. All right, that's a lot. Live stream audience, good to see you. Joe says I live in Codex these days. Amico, thanks for joining from Tokyo, Gabriella joining from Canada, Jose from Santiago, someone here from Los Angeles. All right, so thanks for, thanks for tuning in. If you do have any questions. I don't always do the, you know, live audience stuff as much anymore. It might be a little confusing for a podcast people, but if you are listening on the podcast, FYI we always will have a video version of this on our website at your everyday AI.com because we are going to be doing a demo. Yes, there are going to be some visual things. I'm going to try my best to describe everything so you don't need to watch the video. But if you do, make sure you go check that out. And for our live stream audience, if you do have any questions as we go along, please make sure to let me know and I'll try to tackle them as we go or at the end. One other thing to keep a note, we are going to be doing some sort of, I've changed the name still kind of shooting for a quarter to launch of this. So a cowork Codex cohort. That's a mouthful, right? So if you do want access to that, make sure to repost Today's show on LinkedIn. So if you are listening on the podcast, check your show notes and we always have a link to today's show that we put on LinkedIn. So go repost that and I will give you more information and give you priority access when we finally do launch that Cowork Codex cohort. All right. Woo. That was a lot. So let me get into the facts here. I think there's been this recent shift, you know, probably in quarter one, quarter two, toward everyone running toward Claude, specifically for Claude code and for Claude cowork. And although I would have definitely agreed with you in December, right, when CLAUDE code came out on the desktop, or I think that was sometime in quarter four and in January, right. The combination of Claude code and CLAUDE cowork was unmatched and what it could then do on the desktop and kind of what that meant for knowledge workers everywhere. But Since February, since OpenAI launched a codex Mac app, I've been using Claude on the desktop less and less. Right. I was using it very heavily for a long time, but it's gotten to the point where probably this week I've stopped using CLAUDE code. And I probably think that you should too, and your company should too, unless something drastically changes. All right, now I want to go over three main reasons for that before we go into the live demo and all that. But all right, number one in live stream audience, let me know if you can see my screen. I'm going to be jumping around a lot, lot today, so please let me know, let me make my face a little bit smaller there, see if we can make my desktop a little bit bigger. All right, so reason number one, it's the best model, all right? And there is actually some confusion on this because there's different variants of GPT 5.5. All right? And one of the biggest points of confusion is Arena. So if you look at arena, right, you'll see that GPT 5.5 is actually not doing that well overall. It's the number eight model, right? So for the blind taste test, it's not doing well, but you'll see the version that's tested on arena is GPT 5.5 high. If you go to artificial analysis, which I always say there's three things you should pay attention to, one is Arena. Two is artificial analysis, which is kind of like a conglomerate of all these different benchmarks. And then the, the benchmark GDP val, which is a model's ability front to back to create economically valuable work. So those three things we could talk about. But one thing that's important to know is in arena, they are just using the high version of GBT 5.5. If you look on artificial analysis, they test the extra high or the best version available. And if you look, it is in a league of its Own it is light years better. Yes, three points on this is light years better, trust me. All right, because you have Claude 47 Claude Opus 4. 7 Gemini 1. 3 Gemini 3. 1 Pro, and the old version, GPT 5. 4 Extra High, they're all tied at 57, right? So there was a log jam for a very long time. Now GPT 5.5 high has a 60 on artificial analysis. So reason number one. Well, GPT 5.5 high is so good. I was actually talking about this with someone yesterday. You know, similarly, GPT 5.5 Pro on the web, it's in another league, right? If you need something done correctly, use the Highest version of GPT 5.5. Or if you have, you know, Gemini Deepthink, similarly, it's. It's a whole nother tier. But in terms of Opus 4.7 with adaptive thinking, which I personally hate, the new adaptive thinking on Claude, right? When I want it to think, even I tell it to think. And you have adaptive thinking on, and then it decides not to think. All right, so anyways, reason number one, 5.5 extra high is untouchable. All right? Reason number two, that you should be using Codex much more is, well, Codex's uptime is impeccable, and that's extremely important. All right, so showing here on my screen, something I posted to Twitter and our inner circle community a couple of weeks ago is this uptime. These are from the actual companies themselves, the gold standard, you know, and think about when using this in the enterprise. And this makes sense when you understand how autonomous these systems are becoming. Right? So as an example, Claude has some great features, you know, scheduled tasks inside Claude, cowork routines in Claude code that you can schedule around the clock. Sounds great. And all until what if you have a series of those things going on? Let's just say you have a series of three actions or a series of three routines that depend on each other. And if number one doesn't trigger number two and number three, if they are dependent on each other, all of a sudden become useless. Right? So kind of the gold standard in enterprise software is either three nines or four nines. What that means is we're talking about uptime. So that's either 99.9% or 99.99%. All right, Codex has a 100% uptime, which is absolutely bonkers, right? OpenAI's APIs has four nines, 99.99, and then ChatGPT has 99.8. All right, ready? Ready for this look at Claude Code, 99.2, not 99.8, not 99.9, 99.2. Claude's API, 99.0. All right, and then Claude AI. So Claude of the Web, 98.7, all right? If I'm running an enterprise, I'm not even considering Claude for that one very reason. The uptime can end up honestly costing your company millions of dollars if you lean heavily into, you know, all of the latest and greatest features from anthropic Claude scheduled tasks, inside cowork, Claude routines, all of these things. I, I, I've tried them, I've done them, and I have all of these dependencies. And it got to the point where it is straight up, because even at a 98.7% or a 99.2% uptime, that is legitimately a recipe for disaster for your company. You need to be as close to that three nines as possible. And that's one reason Codex and OpenAI's APIs, I think, are still and will continue to be the bedrock for serious companies that want to do work online. You can't be playing essentially with kids tools if you're doing real work. All right, and then number three, Claude doesn't actually have memory. Codex does. This is that one secret thing that most people don't know about that when I show people, they're like, wait, are you serious? This doesn't make any sense. All right, so apologies in advance because I'm going to have to do some, some bouncing around here. All right, so you might be thinking, of course Claude has memory, right? So I'm going to go into Claude here. I'm going into my, my settings. All right, let me bounce around here. Sorry, live stream audience. There's going to be a lot going on on the screen on today's, on today's show. So let me try to make this easy to follow along. All right, so as an example, if I go into my settings inside Claude on the desktop and I go to capabilities, Anthropic tells you that you can search. And this is on the desktop version, right? So I'm using the desktop version of Claude and the desktop version of Codex here. And I'm going side by side. And this one reason alone should, should be more than, more than what you need, right? Watch, this is going to be the time it doesn't work when I do it live, all right? But you'll see inside, inside Claude, I have the memory options toggled on. So it says search and reference chats, and then it also says generate memory from chat history, all right? Let's try that out. Right. There's a common a thing I learned in journalism school. It says if your mom says she loves you, get it in writing. So never take a setting as truth. All right, so I'm going to tell both Claude desktop. So I'm using Cowork here, but I tried it across the board that my favorite NFL team is the Chicago Bears. All right? So I'm just going to do the same thing here in Claude Cowork and in Codex. All right, simple enough. I'm going to go now to new task, and here's why. So aside from the fact that Claude code technically has three silos, which stinks, right? There's Claude chat, there's Claude code work, and there's Claude code. Those three are siloed. They have no clue what's going on. But unless you test everything out like I do, you have no clue that quad isn't actually telling the truth when there's that toggle inside of their settings that says it has chat history. Don't believe me? Let's find out. So I'm opening a new tab in each. All right, watch. It's going to work this time. So I'm saying, and I'm even telling it, using chat history and memory, what is my favorite NFL team? So in chat number one, all right, I said it's the Chicago Bears. All right, so let me go here. Same thing in Codex. All right, New chat saying, using chat history and memory, what is my favorite NFL team? All right, so here you'll see Claude said something funny here. I don't have access to any prior chat history or memory files in this session that would tell me your favorite NFL team. Nothing about a favorite team is currently visible to me. That's really interesting anthropic, because you have this listed as a capability in your desktop app. All right? I'm not showing you a setting from the web. All right? It says you can search and reference chats and generate memory from chat history. So that alone think of the work that your company does and so many people from using all of these versions on the web for the last year. Let me just say this. Claude's memory on the web is really, really good. So is ChatGPT, but Claude's memory on the Web, I love it. It's great. This is a huge potential pitfall for your company or if you're a solopreneur small business, you know, using Claude across your teams and want to move to the desktop. This is a. Let me just call it what it Is it is a trap. Right. Because we've had this big shift in 2026, going away from prompt engineering to context engineering and everything is about, you know, kind of persistent memory and being able to use all of this information. But right here, before we even get started, before we even get started, I just showed you why you probably shouldn't be using Claude for anything. Right? And here's the result from Codex. All right, so it said your favorite NFL team is the Chicago Bears. Verified from your local Codex chat history. So not only do you have a huge disadvantage inside Claude, the fact that it has three different silos, Chat doesn't know what cowork's doing, coworker doesn't know what Claude's doing. Claude or code's doing. Code doesn't know what's chat doing. But also even within Codex, Codex doesn't know what it's doing within its own environment. Inside Codex, you have one thing, that's it. There's no three silos. And even within that one element, it actually knows what's going on. So when we talk about who's winning the enterprise, let me just. All right, this, I didn't mean to turn this into a hot take, but this is just one of those things that people are so ill informed. Again, what business leader would look at this and say, yeah, this, you know, Claude Desktop sounds like it's for me, just saying, I don't know, if you want to fail, if you want to get left behind, you can do that. Yeah, there's fun things. I still use Claude desktop for a lot of stuff, but when it comes to my serious day to day work, it's all happening within Codex. All right. Yeah. Someone here from YouTube saying that's insane. Had no idea Claude didn't have memory. Yeah, Jackie, memory worked for me with Claude Desktop. So it is only coworking it. Yeah, please, please show me. I, I've yet to see it work on desktop. I've tried it across team and enterprise applications, but yeah, just did a live demo there. Doesn't work even within chat. I don't think they did just have an update. I could try it in chat one more time, but I tried it yesterday, none of it worked. So unless it's pulling, here's the caveat. If it's pulling from your documents, from your Gmail etc, chat will pull things from there. But in terms of if you get something in a chat, the context, whatever, it doesn't know. All right, I spent too long on that. I accidentally went on a little side tangent, but Let me quickly tell you though, the three reasons why I still use Chat GPT for some things. All right, because inside Codex we're going to show you still get access to all the models. You get the image gen, you get the connectors, you get most of what you get inside of Chat GPT. But there are a couple of things I still use ChatGPT for. Number one is workspace agents. All right, so those are great across your team. And even though those agents are built on Codex, I know this is a little confusing. You don't have the same abilities inside Codex to run workspace agents. And one of the biggest things is to be able to run them when Codex is not running. Right. So yes, you have all of these automations and scheduled automations that we'll go over in part two, but Codex does have to be open in order for those to run. So I still have kind of like backup or redundant tendencies that I use inside ChatGPT inside my business account for workspace agents because you don't need to be quote unquote connected for those. Number two, deep research. As far as I know, I haven't looked at the skills this morning. I looked at everything yesterday as I was finishing up planning, but there's no deep research inside Codex now. I'm sure it's something they're going to bring over. And then the third thing that I'm using still Chat GPT for is Canvas mode for quick mockups. Yes, Codex can actually build amazing things, but I just love the Canvas mode. Maybe that's more of a habitat, right? But I love Gemini's canvas. I love ChatGPT's canvas. I love Claude artifacts. So that's one of the reasons why I'm still using ChatGPT. All right, let's look live. Actually, before we do, I need a coffee break, y'. All. Let's. Let's take a second for a quick word from our partners.
- [21:51]
B
Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI Assistant now live in the Adobe Firefly app. The all in one Creative AI studio. Powered by Adobe's Creative Agent, Firefly AI Assistant lets you start with your vision. Just describe what you want and shape the outcome as it takes form with the assistant. The assistant orchestrates multi step workflows, drawing on 60 plus pro grade tools across Adobe Creative Cloud apps including Photoshop, Illustrator, Premiere, Lightroom Express and more to help bring your ideas to life. You can also get started with creative skills, a growing Library of pre built workflows for common creative tasks like batch editing photos, creating mood boards, portrait retouching and creating social variations. Every step the assistant takes is visible, so you can refine, redirect or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI Assistant now in public beta. See it today@FireFly.Adobe.com
- [23:00]
C
all right, let's get back to it. We're going to first do a very quick hands on overview of everything Codex desktop app. So to start with, you get this on a free plan. It's very limited on the free plan, but if you're on the 20 chat GPT account, you have Codex. A lot of people don't understand. They think Codex is a separate subscription. It's not. All right, so you are going to download this. This is a down. This is an app that runs on your desktop. Let me just go ahead and close Claude code here. So it's not taking up any, any space on my computer. So there is a Windows version and a Mac version. There are some certain feature differences a little bit, but nothing too much. All right, so you are going to download Codex, you're going to install it for certain things. You're going to have to give it permissions. The real cool thing is there's a nice onboarding flow where it shows you for things like computer use, you're going to have to go into your accessibility settings on a Mac, etc granted access. All that good stuff. All right, so then you're going to log in to your, you, you log into Codex via your Chat GPT login. So one other thing, if you go into general settings here, all right, there's different setups and there's a new onboarding flow which I really like that OpenAI released a couple of weeks ago where it is essentially built around different Personas. Right. So OpenAI is leaning more heavily into kind of shedding the old Persona, that Codex is a developer tool and it's for anyone. And you'll even see that in Settings. So you can go, you can have a coding setup that gives you a more technical response or control or you can click the everyday work kind of user interface. So if you click that, it literally just looks and for the most part functions like ChatGPT, the old desktop app. All right, a couple other things I'm going to scroll through here. I'm not going through all of these settings. I did want to talk actually. I'll get to that later. All right, let's first go over a simple overview of the desktop app. All right, so like Chat GPT, there's folders and then there's normal chats that don't live inside of a folder. And that's also kind of new. So let me first tell you what a folder or a project, what it's used for and how it's set up. So I'm going to click on the left hand side, there is a normal kind of sidebar. You can toggle it on or off. All right, so if I click New project here, all right, there's a couple things that you're going to look at and be like, what are these things? So, because it runs on your desktop, all right, one of the biggest benefits, but also one of the biggest potential downsides if you don't know what you're doing, is you're granting it access to a folder and it has read write capabilities to your local machine. And that is the biggest. One of the biggest differentiators between using Codex on the desktop, using Codex on the Web, or using ChatGPT on the web is it can literally run and control your entire computer. Right. So with great power comes great responsibility. If you are using this in a work setting, make sure you go through all the right permissions to, you know, say, hey, can I do A, B and C with Codex? Go through all those things. But there's different access that you can have. You can have default permissions, you can have auto review, you can have full access, or you can have a custom configuration. So you can set these up on the project level. And then you're going to give, whoops, you're going to give Codex access to a certain file on your computer. And then I can set that up and then it will be able to read and write on its own. It'll be able to create new folders, new files, pull from information you put in there, etc. So that's the first piece of strategy there because for the most part, you're going to want to sandbox these off, Especially if you're a newer user, what that means is you're probably not going to want to sync this to like a team drive and give it access to an entire team drive. When you're just trying to build something on the side, or if you're just trying to experiment how to automate your work, and then all of a sudden, whoops, it just deleted a file or folder because you told it to, but you didn't know that it was going to do that, etc. Right. So I would always start by sandboxing it and creating a specific folder, right? For me, I have it in my Documents folder, I have a codecs folder. And then underneath there, then I kind of have it sorted out by project. Or let's say if you want to use Codex for something like something simple, which I have a version, I have one of these going where it's just. It's my downloads folder, right? A lot of times I'm downloading a ton of local files to my computer and I want to be able to access them on the go, I want to be able to sort them. I'm like, oh my gosh, what, where is that file? What can I do with it? Etc, right? So that also helps me organize my life, but I also know it could go in there if I'm not paying close attention. And this is with any autonomous agent that runs on your desktop. It can accidentally delete things, move things, rename things, etc. All right, so keep that in mind. So first, when you set up a new project, you're going to go ahead and give it access to a folder. Also GitHub, we don't have to get too technical here, but essentially, if you've Never heard of GitHub, think of it as like a Google Drive. It's like a storage for your code. All right, so in a lot of these instances, even if you're not asking Codex to write code, it will still write code. Right? To accomplish certain things, because it not only has access to your desktop, it can access your terminal. So you can do things with, you know, third party APIs. Essentially, it might be doing a lot of heavy lifting on your computer without you even telling it to. So you don't even have to technically know how a lot of these things work because it has access to your computer through some more advanced features. It can access, you know, your Chrome browser, there's a new codecs Chrome extension, it can access API services via the terminal, all of these more advanced things. So GitHub essentially is an option to, quote, unquote, backup your code to GitHub. And you can do that in a private repo or a public repo. So you don't have to set something up to GitHub. But if you're wondering, like, what does that mean? That's what it means. All right, couple other basic things before we go through our 35 fast facts. All right, so we're not doing a full hands on demo today. I want to keep it simple. Couple things you need to know. You can still choose the different models. So you have low, medium high and extra high, all right? Depending on the. On the plan, you're probably not going to see extra high. If you're on a free plan as an example, you can choose different models. You can choose 5.4, 5.3 as well as 5 point, sorry, 5, 5, 5, 4 or 5.3. So you also have the different models and then essentially different tiers of intelligence. All right? Also, I forgot to notice that I should have said like four reasons you probably shouldn't use Claude for the most part. Limits, right? Your rate limits on a $20 checkbox account are much better than a $100 a month Claude. Not close. On the 200amonth Claude Max plan, you know, you can pretty much use it at will. But if you're even on the 20 or 100amonth Claude plan, you're in, if you're a heavy user, it's not going to work for you, right? Even if you're on the 20 plan on chat GPT, though, you're probably not going to hit limits for the most part, even if you're using extra high a lot. All right, so you choose your model, you can click the plus button, choose your different plugins. You have to connect all of these. So plugins are like apps and certain capabilities and we'll go over that more in part two, there's a plan mode which, well, it helps you plan, all right, which is great. It helps you, you know, if you're about to take on a big project, set up a big automation, it kind of asks you a series of questions to make sure you're doing it right. Then you can upload photos and, or, sorry, upload photos or files. And then you can also click the backslash. So backslash is going to give you some different commands and options that you might not know or you know, you don't have all of these things in Chat GPT. So as an example, there's a feedback button, mcps. You can build MCPS to connect to any data service work tree. You can, you know, personality. They have these new things called pets, right? I'm not going to get too far into it, but click the backslash command. Take some time looking at those things. Also click the AT button. The AT button is how you work with different plugins. So a lot of times Codex will just understand your intent really well. So if I say, as an example, you know, triage my, you know, my, my email, my calendar and my drive, right? So even if I don't say Gmail for email, Even if I don't say Google Calendar for Calendar and even if I don't say Google Drive for Drive, usually it will know those things, but you can still click the AT button when you're typing out your prompt or building different workflows, building different automations, and you can still manually select those apps from the drop down list that you have connected. So similarly, how you would go through and connect apps or connectors in Chat GPT or Claude or Gemini or, you know, co pilot, whatever, the same thing inside of Codex. All right, so that is the basics of using the actual app. And like I said, it looks very much like Chat GPT. You can go down here. There is a chat only section. So if you don't want something to be attached to a project by default, there is kind of a chat section as well. So if you're like, I don't need to give this access to a bunch of, you know, folders or files on my computer, or maybe you just don't want to. Right. So you just have that option down there to go over the chat. All right, that was a lot. So now let's do this. We're going to go through as quickly as possible 35 fast facts to help you better understand and use Codex. All right, so probably not going to be doing, FYI, if you're listening on the podcast, probably not going to be doing a ton of these things on the screen here, but probably I will stop and point out a few of them. All right, so we are going rapid fire here. All right, some of these already went through, so we'll go quick. Number one, Codex is the new super app. All right, so there's been all this news. OpenAI is moving toward this super app. They're kind of putting, you know, Atlas and Chat GPT and Codex into one. Well, it's Codex. So if you're not using Codex now, that's the reason you should be using Codex and you should get used to it right now. Number two, already said this Codex is included with your Chat GPT subscription. You don't have to be paying extra. Number three, important to know there are, there's a web version of Codex and a desktop version of Codex. So I'll actually show you here. You can choose to work locally on Codex on the desktop and there is an option to connect Codex Web so you can kind of, you know, keep those projects or those chats synced if you want to. For me, I don't use Codex on the web a ton. If I do want something quote unquote synced. Normally I'll do it. Normally for me, it's a little Codex, you know, little coding projects, little apps that I'm building. So I'll usually kind of quote, unquote, connect it via GitHub private repo, but you can connect to the desktop version. You can connect the desktop and the web version as well by choosing, you know, changing it from work locally to connecting Codex Web. All right, number four, the desktop state does not sync. Here's what I mean by that. If you use multiple computers, your Codex history is not going to carry over. Same thing with Claude code, Claude cowork, etc. That's the same. So it is to your local computer, which for most people is not going to be a big deal. But for me, right, I use probably about three machines pretty consistently almost every day. So that's one downside for me. So just keep that in mind. All right, number five, and I kind of mentioned this in my intro, Chat GPT works with you, Codex works for you. What do I mean by that? Well, Codex is agentic, right? So the new workspace agents which we went over last week, they're powered by Codex. Codex can use your computer in the background, all right? It can do work for you. It can read and write. That's the big differentiator right there. You know, whereas Chat GPT, as an example, for the most part, can't write. It can read all this information, but you still have to be the one copying and pasting. You know, it can create files, but you still have to save that file and email it, whatever, right? Codex can literally do everything. Read and write, you know, browse the web sites that you're logged into, it can do it all. Number six, Codex is the unified workbench. What do I mean by that? Unlike Claude Desktop that has three different silos that don't know each other, exists with chat code and cowork, Codex is one area. You don't have to, you don't have to make any concessions and say, okay, well, I'm going to work on this project and, you know, it's, it's going to have to live here and, you know, these other areas aren't going to know about it. Well, you don't have that problem with Codex, let alone for the fact that anthropic memory doesn't actually work on the desktop. Right, And Codex does. All right, number seven already mentioned this. GitHub is essentially like your code's Google Drive. So even if you're not writing code, you may still be technically Having a code base when you're working on something and you can sync that directly via GitHub. Also number eight, the GitHub connection is by default. So once you set it up, once you don't even, you know, there's different settings inside Codex to where you don't even have to, you know, do a pr, you know, push something live, you can have it do it automatically. So that's really cool. Just like in Google Docs as an example, it's going to save automatically. You do have that option if you are trying to do actual building inside via Codex, that that connection is there by default. Number nine, MCP support comes built in. This is one thing that I absolutely love. So unlike certain versions inside ChatGPT. So as an example, ChatGPT has MCP connections, but only on team accounts, which is kind of stinks for me, right? I love using my pro, my 200amonth pro account the most, but there's no MCP support in the Pro account. So inside Codex, it doesn't matter what paid plan you're on, you have MCP support. What does that mean? Even if there's not an app built in kind of in the app store for Codex, you can just build it with mcp, right? So my great easy example is we use Beehive for our email newsletter. There's not a Beehive app inside of Codex, but there is an MCP model context Protocol server. And so I can essentially do anything. I could literally run my entire newsletter business inside of Codex because there is that read write capability through the MCP support. All right, number 10, and we're going to be going over this more in part two. There is plugins, skills and automations. So essentially skills are an open source standard. So if you use skills inside Claude or Gemini, you can actually use those same skills and you can actually import them. I'm going to get to that here in a second. But it has default access for plugins, skills and automation. So skills kind of teach Codex workflows. Plugins connect codecs to your external services that you use via like a traditional app or connector. And then automations let Codex repeat tasks on a schedule. So you can run those manually, you can set them up to run at a certain time, every single day, once a week, etc. All right, number 11, kind of related to what I just said, importable Codex workflows. So Codex, or I guess OpenAI just released this a couple of weeks ago and it's part of the onboarding system now, which I really like. But you can also do this at any time in your general settings is you can import literally everything, all of your skills, your chat history, your memory from any other provider. So if you are a Heavy Claude code, Claude cowork, user, heavy Gemini user. If you're using, you know, skills as an example, Inside Copilot, you can export all of those and then import them into Codex. All right, number 12, which I probably won't even get to too much in Part two because it is a little more niche and it does require a higher tiered plan. But Codex has something called Chronicle that's absolutely bonkers. So if you remember, Windows Copilot announced this feature called Recall or wait now, now I'm forgetting. What's it called? Recall or Rewind. This is one of those things I, I talk about AI so much, I forget Recall. All right, so Microsoft Copilot announced this thing called Recall. There was all this security concern and all this hoopla, so they ended up delaying it like a ton of times. But it's actually a built in feature inside Codex. So all that is is Codex continually takes essentially screenshots of what you're doing on your computer. So when you have that feature enabled. So it's almost like a visual memory for anything you've done on your computer. So pretty cool. There's obviously some security considerations there, so keep that in mind. All right, number 13, you supervise agents and Codex, you just don't prop. Like I said, a lot of the time I spent in Codex is more I, I've used this analogy before, kind of the, the burger analogy, right? Front end and back end, the bun. Spending more time as the bun and less time as the meat doing the actual work. So much more time I spend inside COD is refining my agents, refining my schedules, giving them more and better context, looking at the chain of thought and improving it. So I'm really just spending more time on the front end and then on the back end taste making or reviewing or approving. So inside Codex, I'm spending less and less time doing the actual work and more time setting up orchestration on the front end and then doing the good old expert driven loop on the back end and having those iterations. All right, number 14, this is a good one. Steer versus run. Let me just do a quick, quick, a quick little example. So I'm going to say, I'm going to say solve all my problems. All right. Inside Codex, you know, take your time. All right. This isn't an actual prompt. I'm just doing this to show you an example of Steer versus Q. So this is important. So this is something inside Chat GPT. Maybe you've run into this because only on the pro model tier can you interrupt Chat GPT. So let's say normally it takes two to three minutes and you're like, oh man, I should have said something else. Do I stop? Do I wait? So inside Codex, at any point you can send another message and either have it wait in the queue or steer. All right, so in this little example, I'm just saying solve all my problems, take your time and then I'm going to say work only problems. All right, As a steer. So as an example, all right, I'm just telling Codex to solve all my problems, take your time. And it would probably take a lot of time. And then I'm going to say work only problems. I'm going to click enter there. And then you have this option to steer instead. So even if you're working on a complex task, Maybe it's taking 2, 3, 4, 10 minutes having that option to either put something in queue, which is a hack in and of itself. I do that all the time. I'll probably touch on, touch on that a little bit more in part two. But you can also steer any single query while you wait. All right, number 15, fast facts. Terminal awareness enables iteration. So what this means Codex can read and understand failures whether you're doing something technical or non technical. You no longer have to explain what went wrong because Codex can literally read the terminal and see what went wrong. That's huge. Speaking of that, number 16, visual verification changes everything, right? So in the same way that Claude Cloud has a great preview inspector, I think Codex has a better one, right, because it literally has an in app browser where it can. Let's just say you're building an app, building a web app, building something, right? Not only can it test it and render it in real time via its kind of built in file browser in built in web browser, but it can also do different viewports, which is absolutely insane. So I don't know, let's just say you're building a website in Codex, so not only can you see it rendered in real time because it can obviously write and render code, but you can also look at different viewports, which is absolutely bonkers, right? So you can see. Show me what this looks like on a desktop, show me what this looks like on a mobile view, on an iPad view, right? Different viewports. It can actually preview your design and it can actually do all that automatically without you having to do anything. All right, number 17 browser, Chrome and computer use, that is huge. So Codex will automatically understand what is the safest tool or the fastest, most efficient, most accurate tool to use first. Because there's so many different ways that it can, in theory, get work done on your desktop, right? It can control your actual computer, it can control your browser via computer use, it has a new Chrome extension, and it can also do things via the API. So it's actually super smart in understanding how it should attack a certain property. Them number 18. This is big. I did talk about this a couple of times on previous shows, but it has separate mouse and keyboard control. This is one of the biggest downsides of, you know, Claude's computer use. When it first came out, it went mega viral and I was trying to use it and I was super excited and I'm like, this thing just stinks. Because what happens is when it first came out, it mouse jacks, it keyboard jacks, it screen jacks. What that means is when Claude computer use first came out, if you were using a browser, it took it over and you couldn't use it, it took over your mouse. Right? So essentially, Claude's computer use, more of a gimmick, I would say. Or it's more of something that works overnight while you're not working. And good luck on a 20amonth plan getting computer used to do anything. You do have to be on a max plan to get any actual utility out of it, unless you're doing a single run a week and that's all you're using your plan for. All right. The cool thing about Codex, it literally has its own. So it, it works via the user interface, but it also works via the operating system level. So what that means is it technically has its own mouse and it has its own computer input. So I think Codex, with its browser use in, sorry, computer use is probably the first truly autonomous coworker, because even with Claude cowork, it kind of takes over what you're trying to work on. So it can really only work for you while you sleep or while you're not using that computer. With Codex, I'm all the time working at the same time, which is completely, I think, changes the human connection to AI. To say it changes what's possible is an understatement. All right, number 19, slash side. All right, so what that is, you can just click slash side. So I'll give you an example here. You can just click slash side and start chatting inside Codex. What that means I actually have to have a chat going to use that. All right, so at any time you can just do chat slash side and essentially you start a side chat about anything that's happening in the main chat without taking your main chat off the rails, if that makes sense. All right, so pretty cool feature that, that most people don't know about. Number 20 sub agents. My gosh, you can just literally tell Codex to create a bunch of sub agents. You can give them roles, give them tasks, right? Have this sub agent go plan, have this sub agent go research, have this sub agent build and have this sub agent do qa. Or you can just say go use three sub agents, break the task off how you want and then get back to me once your sub agents are done. So you don't even have to be technical, you can just go have it spin sub agents. Yes, you will go through your your limits a little bit faster using sub agents. But regardless it is a something you have to know. 21 work trees isolate parallel work. What does that mean? Well, work trees without getting too technical essentially let different tasks happen without smashing into each other. So this is where you can have multiple agents running safer and more manageable on your local machine. So not just sub agents, but you can obviously have, you know, 10 different tasks going on at any given time inside Codex and there will be kind of a visual cue here. Let me just do that thing I did earlier. I'm going to say solve all my problems, take your time. Right, just so we can see. So anytime a chat is working, you'll see here on my screen there's a little toggle that shows that it's working. So work trees allow you to have multiple agents, multiple tasks, multiple automations, multiple big projects going on at the same time and they're not going to run into each other essentially. All right, number 22 slash goal for power users. We're going to tackle this more in number two. But Gold is amazing. So this is a newer update actually Anthropic just came out with this after they saw how popular it was with Codex. And that just allows for long running work with a clear done condition. 23 automations keep checking things. So setting up these automations we will go over them in number two. I had to clean up my account because I didn't want to give away too much, you know, private stuff here on my screen. But automations can run around the clock. Number 24 rules make Codex yours. So there are rules and custom instructions that you can set up whether it's at the project level across your whole account, etc in the same way that you have that level of customization in quad code in, you know, chat GPT on the web. You have the same in Codex. Here's a cool one. Number 25 hold to dictate. This is a feature that OpenAI came out with in Codex a couple of weeks ago. I think it's one of the more underused features, but it has built in system wide dictation. So you don't even need a separate dictation tool. It's literally built into Codex. Even if you don't need to use it for codecs, you can just use it for anything on your computer. All right, number 26, here's another cheat code. There is a, a model called Spark. You're probably not going to find it because you have to go under this other models tab and it's only Codex 5.3 spark. Although OpenAI has alluded to that there will be a future version of Spark. Maybe it'll be 54, maybe it'll be 55. We'll see. Spark is essentially a version, without getting too technical, of OpenAI's models that have been trained on the Cerebras chips. All that means is it's crazy fast. It's not even fair using computer use. With 5,3 Codex Spark I, I did some side by side testing. It was 18x faster than Claude codes computer use. That's not an exaggeration. I literally timed it 18x faster. All right, it was about 5 to 6x faster using the normal 5.5 with computer use. But 5.3 codec spark, it's not the, you know, the smartest model in the world. But if you need basic computer use, that Spark model is absolutely bonkers. 27 non technical Codex onboarding already talked about this, but the UI is moving away from, you know, being this terminal only developer tool. And if you are using codecs for the first time, it has a great way to walk you through non technical Codex users, right? So being able to understand projects, plugs in, plugins, Automation, browser, voice, etc, 28, this is huge. Doc, slides, spreadsheets, these are all built in skills. So Codex, much like Chat GPT or CLAUDE can do on the web, it can just build, it can just build you things, right? It can just build you a document, it can build you a presentation. You don't have to do anything special. It's a built in skill. All right, number 29, messy tasks make the best demos. What do I mean by that? A lot of times the best thing to do inside of Codex or sometimes I start this in chat GBT is saying Based on everything you know about me? Well, first go look at Codex. Then, based on everything you know about me, what are some ways I could start automating my work inside of Codex and then start building those things in Codex and you can build them all in natural language, but a lot of times it's just starting out very messy and then refining over time that you're going to get your biggest wins inside of codex. Number 30. Codex means validated work. What do I mean by that? Well, geez, I mean, Codex is so valuable because it cannot just do the thing, but it can inspect the result, it can fix issues and explain to you in natural language what has changed. So, like I said, it's almost like having multiple AI models working at the same time, because it can actually see, understand and validate its work. Number 31, the pace is moving fast. All right, so Tebow, who is the head of codex there at OpenAI, just confirmed, I think last night, Codex is moving to a weekly updating schedule. I think they're going to have Tuesday, Wednesday, Friday updates. The big or no Wednesday, Thursday, Friday, I think. Or Tuesday, Thursday, Friday, I already forgot. Thursdays are going to be the big releases and then there's going to be some smaller, you know, Polish bug squashing. Right. But essentially every Thursday there's going to be a big release on Codex. This isn't one of those things that a project that's going to be abandoned. Codex is getting more and more powerful by the week. All right, last but not least, let's go. Let's go through it quick because I already covered these last couple fast facts. 32. Claude splits chat code and cowork. They work in silos. They don't know the other exists. That's not the case with Claude, or, sorry, with Codex number 33. Claude. Don't, don't get me wrong, Claude has strong work metaphors. So what do I mean by that? Live artifacts, Amazing. Dispatch, amazing. So, you know, one thing I do want you to know about Codex is it doesn't mean it's the absolute best at everything. Although I do think it is hands down the best platform if you only choose one. So keep in mind there are still some features in Claude that are a little bit better. Although I do know that OpenAI has alluded to the fact that they're bringing kind of mobile control, which is kind of the dispatch version in Claude. All right, number 34. Codex has that execution energy. What do I mean by that? Codex feels like a legit agent that can do anything that you tell it to as long as it has the right access to your data. One of the biggest issues or one of the biggest obstacles I've had to overcome with Codex is understanding its own limitations. Because every time I tell it something and I'll work through the problem, eventually I'll be like, oh, Codex could actually do this. I just thought it couldn't. I didn't have it right. I didn't share enough context, I didn't iterate enough. Right? So just the execution energy of Codex is unlike anything I've seen. Because like I said, a lot of times in Claude desktop you might have to go straight to Claude code for something which can be, number one, extremely expensive, extremely slow, or it can just be more limiting because there's certain things that you might want to do carry over from chat to cowork to code that you just can't do. 35 safety makes agents usable. So built in, although you can skip these things, I wouldn't recommend skipping these things for new users, but sandboxing approvals, protected paths, network controls, browser permissions, all of those things are built in by default. All right, so that was our last 35 fast facts. My gosh, that was a lot. All right, I said I was going to break this into two parts. Imagine if I did this in one part. But I cannot emphasize enough as we wrap up today's show. If I miss your comment, I'll get, I'll get to it after the show here. So I want to say this. Start using Codex today. I don't care if you're tied down to Chat GPT. I don't care if you, you know, feel in your mind, oh my gosh, Claude Code and Claude Cowork are amazing, right? Codex is all of those things. It is the best of Chat GPT minus a few things, right? So I'll say It's this. It's 95 of what chat GBT can do. It's 95% of what Claude Cowork and Claude Code can do, all and more, all in one unified interface. And the best thing is over the past couple of weeks we've seen a, a specific reach by OpenAI to make Codex more and more accessible, more and more understandable and just more and more useful to non technical users. So forget about the word code, forget about anything, start using Codex. I'm telling you, it's, it's, it has the largest percentage of usage for me that any AI tool has ever have, right? I'm using all of the big four, right? Usually every single day. And kodak has the largest share that any single platform has ever had. It is that good, you need to start using it. All right, so as I as I wrap if this one was helpful, please remember to repost this. Check out the show notes, click that repost button, then go to your everyday AI.com thank you for tuning in. Hope to see you back tomorrow and every day for more Everyday AI. Thanks y'. All.
- [56:33]
B
Meet Firefly AI Assistant now live in Adobe Firefly, the all in one Creative AI Studio. Just describe what you want to create in your own words and the Assistant handles the rest, orchestrating multi step workflows across Adobe Creative Cloud apps including Photoshop, Premiere Express, and more. In one conversational interface, you direct the outcome while the Assistant accelerates execution. Stay in control with the ability to step in and refine at any time. See it today@firefly.adobe.com.
- [57:08]
A
And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going For a little more AI magic. Visit your everydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.