transcript

Speaker 1:
[00:01] 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.

Speaker 2:
[00:17] Maybe you noticed the OpenAI updates to Codex in the middle of last week, but chances are you either didn't see them, you didn't really test them, or maybe you forgot about them already because in the AI space you assume feature updates like these are a dime a dozen. But in this case, that assumption would be a gargantuan mistake. Why? Well, I think that what OpenAI released in its updated Codex app actually represents one of the largest changes in how AI works that we've seen over the past three years that I've been covering it daily. And if you listen to this show, you know I don't say things like that lightly. So why is this random Codex update potentially one of the biggest shifts in the modern AI era? It's not because that this is an official preview of OpenAI's forthcoming super app or it's kind of forging three of its different desktop apps into one. No, it's more about what this update represents. This is the first true autonomous co-worker that you can actually work with. So, it is our weekly Putting AI at Work on Wednesday show and we're going hands on with what will likely be the future version of Chat GPT and the latest Codex release from OpenAI. So before we get started, here's the big picture. Chat GPT is going through a huge change and you can get a sneak preview of the future super app in today's Codex. And overall, if we look at the bigger picture about what's going on, it's undeniable. Both the consumer and the enterprise momentum is moving to autonomous operations on the desktop, right? Whether you want to talk about things like OpenClaw, one of the most popular open source, well, the most popular open source piece of software ever, or more enterprise offerings like what you have from Claude. Everything's moving toward the desktop. Speaking of Claude and Anthropic, they actually jumped off to a big early lead in that very category, both with their Claude code when they released it for the desktop and what it ultimately led to. Well, Anthropic noticed that people were using its Claude code to do non-technical things, which led them to create Claude Cowork, right? Now we've seen a whole bunch of variations of this Cowork from Microsoft with their Copilot Cowork, all the way to Google's upcoming rumored agents, their updated agents. But what we've seen is this shift to desktop autonomous work. And Chat GPT has been building their Super App in plain sight to take advantage of those actual things. And they aren't saying, here's a preview of the Super App. No, they've just been rolling it out into Codex. And the cool thing is, is you can start using it today. The future version of Chat GPT. If you can just get over the word code in Codex and stick with me as I show you how non-technical business leaders like you and me can use it. And all you really need is a paid Chat GPT plan. Nothing else. So on today's show, stick with me for the next probably 30 minutes, because we're going to go hands on and that one could get crazy. All right. But here's what you're going to learn. You're going to learn how to use Codex and why the traditional Chat GPT interface may change forever. You're going to know why Claude is actually way behind in the desktop race by being first and also really good. You're going to know how to leverage OpenAI's version of their desktop software. And you're also going to know the five essential steps to go from Codex beginner to novice in no time. Sound good? Sound like a plan? All right. Let's get after it. Welcome to Everyday AI. My name is Jordan Wilson. If you're new here, this thing is an unedited, unscripted, live, daily, live stream podcast in newsletter helping everyday business leaders like you and me keep up with the non-stop barrage of AI updates coming at us. We just got to duck our heads. Don't worry. I spend literally my entire day every day seeing what gets thrown at the wall, what sticks, what's worth it, what's not. I tell you how to use it. All right. So if that's what you're trying to do, make sense of everything that's going on to grow your company or career. It starts here, but make sure you go to youreverydayai.com, sign up for the free daily newsletter. We're going to be recapping the highlights from today's show, as well as keeping you up to date with all of the other AI news and updates that you need to know to be the smartest person in AI in your department or company. All right. Did you guys know that there's this new super app thing, right? OpenAI, after months of rumors, they did confirm it recently that they are working on a super app, but it's not out yet, but they essentially confirm that, well, what they're rolling out in previewing in Codex is the foundation for their super app. So make sure if this episode is helpful, we're going to eventually do a co-working type, cohort or a co-working class, right? Where we go over Claude co-work, but we're also going to be going over Codex or whatever it's called at the time. So if you do want early access to that, it's going to be free. I don't know if it's going to be a live cohort or a recorded class, but just make sure to repost this show on LinkedIn. So if you are listening on the podcast, thank you. We always put the link to today's show via the LinkedIn post. So make sure to go repost that and I'll make sure that you get early access to that cohort whenever we do launch it. Today is going to be slightly less hands on, but we're still going to do it at the end and it could go terribly wrong, but we'll see. So not only do I think this new Codex update changes what's possible, but I also think it changes who can use it simply. And what I mean by that, I think, well, I've been very vocal about how good Codex is since the app came out in February. My gosh, it seems like it's been out for years. It's running on my computer. If my computer is open 24 hours, it's running 24 hours a day. But I do think you had to have a baseline technical understanding. Not of code, but you had to be comfortable around things like GitHub and work trees. But I think now we're past that with some of these recent updates. So that really changes who can use it and who can take advantage of this today. So first, let's go over what is actually new. So yeah, to hit rewind, if you have a paid ChatGPT plan, that's all you need. So you download this. This is a desktop program. All right, it's different from their Atlas browser. It's different from the ChatGPT app. It is Codex app. And the big difference is, well, when you're setting it up and you go through permissions, we're not gonna go through that a lot in depth today, but you essentially set it up to where it can both read and write certain folders on your desktop. You give it access. And depending on if you're on Mac or Windows, there's different accessibility settings that you enable where it can go perform actions much like a human would. So here's what's new with the update that Codex, or sorry, that OpenAI just rolled out to the Codex app. So now it essentially, to simplify it, it works at the operating system level, not the user interface level. So you can still see what it's doing on the user interface, but it actually works in the background. And that part right there is huge. We're gonna be explaining that a little bit and how that's different from other computer using agents like computer use in Claude Code or computer use in Claude Cowork, but that is the big step change here. And what goes along with that is it has its own independent cursor on Mac, right? So you can literally see that and we'll see in the demo and that part is huge as well. You can also with those new changes, you can also have multiple agents that run simultaneously. It doesn't screen jack or interrupt your work. And that's why I say this is the first true agent that you can literally work with, right? And that part I'm going to touch on a little bit later as well. Whereas Claude's computer use requires you to kind of stop working because it takes over whatever you're doing. So some new, some other new features in Codex. Well, it now has access to the new Images 2 model, all right, that OpenAI released on Tuesday. There's an integrated web browser that lives inside of Codex, as well as an integrated file viewer. Those two things are huge. We're not going to be touching as much on the technical side, right? But in terms of previewing, debugging code, different web apps that you're working on, whatever, to be able to do that inside of an actual in-app browser is huge. The same thing with having an in-app file browser. It just makes things more reliable and faster. And ultimately, that makes things more token efficient, right? Also, there's 90 plus new plugins including Atlas and GitLab. There's all this usual Gmail, Slack, all that good stuff. And then Chronicle, which isn't always on memory. I do believe you have to be on the pro plan, the $200 a month pro plan for that one. We'll see if they roll it out later. But essentially, this is a version of Windows Recall. More on that later. But yeah, a very viral Windows Recall that was under scrutiny for 18 months. Yeah, Codex just literally rolled it out. And I don't think anyone really noticed. All right, also what's new and updated is the memory and automation. So there's now persistent memory that remembers preferences and workflows across sessions. All right, Codex can also schedule future work across days or weeks. All right, that part's not necessarily new, but it will suggest sometimes scheduled tasks for you because now it will proactively suggest automations based on previous actions. So if it sees that you're doing the same thing manually over and over, it's essentially going to give you a nudge in being like, hey, why don't you set this up as an automation? Here it is for you. Do you want to run it? So not only do we actually have a true autonomous co-worker in the computer use, but we actually have a proactive assistant that is suggesting things when it sees us working too much manually over and over, right? It's when you're hitting your head against the wall and someone's like, hey, maybe you should stop hitting your head against the wall and do this instead, right? That's essentially what we have. Availability, like I said, it's available right now on Mac and Windows desktop. Some features are not yet available for Windows users, but most are. It does require any paid ChatGPT plan. That part's important because this might have already been squashed, but like late Tuesday, so last night, there's all this back and forth of Anthropic may have removed Claude code from their base pay to plan. All right, so we'll see if they actually get that solved. It'll be in today's newsletter, so it looks like it's still working itself out. But regardless, OpenAI's leadership came on and said, hey, you're always going to have access to ChatGPT, or sorry, to Codex on any paid plan. All right, so now let's go through some of these features a little bit more in-depth and set the context here because yes, this is our AI at work on Wednesdays. I tried to describe and show you in-depth one feature, but I think that you have to understand the context and why this is important because I'm guessing the majority of people listening to this, you're saying, okay, I use ChatGPT in the browser, right? I tried the ChatGPT desktop app. It wasn't that much better. It wasn't that much different. Like I'm good, right? Or maybe I'm using Claude code or Claude code work and I love it, but you have to understand here, this is where the puck is headed with the super app. So Thibaut, the head of Codex, said in a recent interview, so I'm quoting this from an in-gadget article. So he says, we're building the super app out in the open. The release is about developers. In the future, we will broaden it up to a wider audience until then, and that's end quote, until then the latest version of Codex offers developers multi-purpose AI agents that can work across a larger surface area while being more proactive. In practice, that translates to a host of new capabilities starting with computer use. All right. Here's what that means. You essentially have these three different desktop apps. And until Codex came along, I was using all three of them every day. The ChatGPT desktop app, which was not very different than the web version, there were some different things. You used to be able to quickly give ChatGPT access to different apps that were running on your computer. It had kind of like a meetings mode, but some of those things got phased out. So the actual ChatGPT app wasn't that different than what you could access on the web. Then they had their Atlas app that is their agentic web browser, which has been really great. And then you had the Codex. So Codex is actually a year old, but the desktop version just came out in February. And now, yeah, the stats already outdated here, moves quickly. So now, OpenAI says that it's already used by 4 million people right now. But that is the past. But now, like I said, they are treating Codex as at least a preview of the future super app. Who knows, maybe that super app will be called Codex. Maybe it will be called something else. I hope it's called something else, because I think people look at the word code and they're like, it's not for me, trust me, it is. But here is the big takeaway that I cannot emphasize enough. This is the shift from the user interface level, interface to the operating system level autonomy. Because most computer using agents, whether you're talking about ClaudeCode work, ClaudeCode, there's a dozen or so popular desktop computer using agents. For the most part, they all run the exact same. They literally screen jack your computer. So what that means, you can be working on something, and I've complained about this for weeks, right? Probably months on the show. You can be working on one thing, try to work, co-work with Claude. And well, you can't co-work with Claude because it would take over your mouse. It would take over your screen, your browser tab, whatever it was you're trying to work on. So although, yes, very impressive, very viral moments for computer use, but it wasn't a true autonomous co-worker. It was just an autonomous agent or a semi-autonomous agent that needed your actual computer. But now with the Codex update, you have that background parallelism. So it operates natively within the Mac OS operating system. So it understands the user interface, but it actually works. Yes, it does. It can work across the user interface, but it can simultaneously work at the operating system level. So literally multiple cursors, multiple agents working on your computer while you're working on it as well. So it can both test front-end app changes silently. It can use in-app web browser and file browser to test those things. It can create images via the new images to feature. Then if you are coding, it can review PR simultaneously. AI moves too fast to follow, but you're expected to keep up. Otherwise, your career or company might lag behind while AI native competitors leap ahead. But you don't have 10 hours a day to understand it all. That's what I do for you. But after 700 plus episodes of Everyday AI, the most common questions I get is, where do I start? That's why we created the Start Here series, an ongoing podcast series of more than a dozen episodes you can listen to in order. It covers the AI basics for beginners and sharpens the skills of AI champions pushing their companies forward. In the ongoing series, we explain complex trends in simple language that you can turn into action. There's three ways to jump in. Number one, go scroll back to the first one in episode 691. Number two, tap the link in your show notes at any time for the Start Here series. Or you can just go to starthereseries.com, which also gives you free access to our inner circle community, where you can connect with other business leaders doing the same. The Start Here series will slow down the pace of AI so you can get ahead. So, and that's why it's a big deal. Because whether you're looking at something like OpenClaw, right? OpenClaw, if you're running it safely and using it the right way, it should have its own computer. But then it seems like you're still constantly having to monitor it. Claude Code, I already talked about the downsides. I love Claude Code, Claude Cowork. I can't use it when I'm working. So for me, it wasn't a true autonomous coworker. It was a worker that sat at my desk, right? And I didn't like that, you know, because the only time Claude Cowork could do any work for me is overnight. And right at that point, you kind of risk things. And that's why I think it's important to dissect this a little bit, because things come up all the time, right? And things obviously still come up in Codex. But the difference is when you can share your computer and share your space with this autonomous agent, if something comes up, you don't have to stop what you're doing and you see it, right? So many times, right? I would think I'd be kicking off, oh, here's a great session on Claude Cowork on Claude Cowork that hopefully it can get a lot of work done overnight. And sure enough, 20, 30 minutes in, I'm asleep, right? So it's in the other room. And sure enough, an error pops up or a permission, even though I'm on dangerously skip permissions, right? Which is a bummer. So, I mean, you still could run into that with Codex, right? In terms of trying to get it to do, you know, long range work overnight or something like that. But this is something you can literally have someone else working with you. And when you do need to check on something or approve something, you can't. So in terms of a true unlock of consistent productivity, this is the first true autonomous AI agent that I think everyday people can benefit from. All right. And then we didn't even get into the the Windows recall feature that is essentially just chronicle. So let me just quickly touch on this. We're not going to actually demo this because it's a little, I don't want to be accidentally doxing people, right? So I'm very careful on the live demos that I do. So the chronic chronicle is a research preview, but I'm reading the definition here. So chronicle is an experimental feature that augments memories with context from your screen with chronicle enabled. Codex references what you've seen to provide more helpful contextual responses to prompts like finish what I was doing or update this dashboard. There's obviously ongoing costs in terms of your kind of rate and it does cost more because it's essentially taking persistent screenshots and committing those things to memory, which is both really cool, but there's obviously privacy concerns around that. So I wouldn't recommend that unless you, you know, if you do work in a larger organization and you've gotten Codex approved, you might want to run that one up the food chain. If you do have approval to use it, you know, obviously use your own best caution if you are a business owner, entrepreneur or whatever. But regardless, that is a huge unlock. All right. Now, I'm going to quickly go over the five essential steps to get the most out of these Codex updates before we go live and do a demo. All right. So number one, connect your tools before you do anything else. Right. So whether that's Slack, Gmail, Notion, 90 others, right, there's custom MCPs, which are great, as well as computer use. And when you combine computer use, you'll see, well, you can essentially any other desktop that, or sorry, any other program that you have downloaded on your desktop, it essentially looks like and functions like an actual app. Right. So it's not just like this loose computer use. It's strange. I don't know what kind of witchcraft the team at OpenAI had brewing in the pot for this one, but even just the use, how this computer use interacts with your computer and interfaces with it, I don't understand the technology. It is mind-baffling. It's just kind of fun to watch. All right. So the reason why this is the first step is, you know this, if you've taken our free Primepump Polish course, if you've heard me talk, context over everything, right? Not just context, but committing that to memory, cross-chat memory, things like that, chat history, right? But the more context from day one means smarter suggestions immediately and the ability to schedule things. Number two, port your top three Claude code work prompts and run them side by side, right? Don't take my word for it, just because I've done my own testing, right? I say this, don't believe benchmarks, run your own use cases. Find the true value, maybe three things that you are getting great results out of a co-work or Claude code and maybe three things that you haven't been able to run those things side by side in Codex to see if it's going to be the best for your needs. Number three, audit every single workflow that requires clicking or browsing and test in Codex. Here's, I don't think people realize just the amount of mundane tasks, right? And here's a huge cheat code. I don't know if I've ever shared this secret in 750 plus episodes. One thing I love doing this, doing this or how I can do this is I record a quick video. I'm very descriptive in what I'm doing. And then I upload that to Google's AI Studio. You can use one of the flash models, right? The thing is Google's AI Studio can see video and say, turn this into a very descriptive step by step tutorial according to both what is said in the transcript and what you see on screen, right? Share with it, you know, codex or Claude code or whatever you're using. And it's going to write you up a great little manual, FYI, but Claude's computer use, as an example, locks you out of your own machine, right? Because you can't really use it. Codex runs in the background. So you need to audit anything that you do that takes up time, right? If you do the same little process, right? A bunch of clicking, context dragging, 30 minutes every single day, that is one of the first things I would look at. Is it low risk, high scalability? Those are the things if you're doing them routinely, you should audit that workflow. Number four, let memory replace your setup time, right? Stop re-explaining your preferences every session. So Codex does and can learn your text act and workflows once and then carry it forward automatically. Like I said earlier, the combination of that persistent memory plus Chronicle plus suggested automations is, I think, one of the stickiest combos for power users. I've been loving it so far. And then last but not least, schedule one recurring task after iterating. I say this all the time, you should be studying the chain of thought. I'm going to show you here a way to get a little bit more context on how to use ChatGPT, but you need to understand how these systems work. And sometimes they give you more context in one app versus another, right? You're going to get way more context on how GPT-54 Pro works inside of ChatGPT versus how it works inside of Codex. So I'm going to show you some tricks as we do this live, but you should always be iterating because generative AI is generative AI. I've run the exact same prompts verbatim two or three different times in Codex. More on the general side and getting very different results. So by doing it multiple times and really trying to look as much chain of thought as possible, you can see where things may or may not go wrong and then make adjustments in your prompt. In doing that, all you're doing is so you can then set a schedule and I'll show you how to do that as well. All right. This y'all, I've done hundreds of live demos. This one might be the trickiest one. So live stream audience, bear with me. Podcast audience, you can always go to our website at youreverydayai.com. We have all the video versions up there. So yeah, hopefully this works. Live stream audience, let me know if you can see my screen. Let's see if we can get this to work. All right. So there's a very good chance this stream could cut out. We'll see. Because I am gonna use computer use and I'm gonna tell it to launch a new tab. So I hope that it's gonna work on the tab that I have open. But I don't know if I can describe this, but to switch over to share that part of my screen, I have to click on, you know, I'm using StreamYard. So it's usually gonna see the active window and then open a new tab. So hopefully I don't lose everyone in the middle of this, but let's go for it. All right. So I will quickly describe kind of this prompt that I'm running. All right. And I do have it up here on my screen. And apologies, I'm gonna read it all, but I'll try to go fast. And you'll see for our live stream audience, there's certain things that are lit up, right? So essentially, you know, there's all kinds of different commands in the interface. And I'm gonna go over this interface later. But essentially, when I click the app button, I can at tag or call on these different tools or apps. So that's the things that are kind of colored in. That's what I'm calling to those apps for. So here's what we're doing. And this is again, I want you don't get caught up in the weeds of my very specific use case. I want you, as I go through this, to think of your specific use case. Where can you pull together all of these different apps and tools that you use, combine them with computer use, combine them with a super smart brain, and start to schedule and repeat some of these workflows? In context, I am using the model GPT-53 Codex Spark. I love it, especially for computer use because it is faster. So it is not nearly the most powerful model. I would probably get way better results if I did not use the Spark model, but it is faster. And I kind of do want you guys to see how fast computer use is because computer use on Claude is slow, slow, slow, slow, slow. Doing simple things, you know, like going to a website, finding something on a drop-down menu, navigating to one page over, and summarizing something might take like 10 minutes, right? So we will see how long a very drawn out kind of prompt here goes. So here is what I am saying. Please navigate to Twitter in the open browser using computer use. Click on the home tab and find five AI stories that are trending. Don't do any searches or filtering, just see what people are talking about on the home screen and scroll the bit. Then using my connected Gmail app, find today's Everyday AI newsletter sent from info at youreverydayai.com and see if any of those, I should probably say those AI news stories, were covered in that day's newsletter. And I should say in today's newsletter. Find the one most interesting topic based, actually I don't know if today's would work in, yeah, in that day's newsletter. Find the one most interesting topic based on what you know about Everyday AI and what our audience may want to know. Again, I'm using my memory persistent, both saved and persistent memory. Then I'm saying then with computer use, open the Claude app on my desktop. All right, so you'll see here, Claude is showing up as if it was a connected app, but it's actually not, there's no Claude app. It's showing up like that because it's because computer use, it sees that it's an app, right? So I don't even know how that works. It's amazing. Anyways, open the Claude app on my desktop. Do not use the Claude Chrome extension. You must launch the connected Claude app on my desktop. Let me just do that one more time. There we go. The Claude app. Where did it go? Claude app on my desktop. Sorry. Once the app opens, search for more information on that story you selected, making sure to specify information from today's date. You should instruct it. You should instruct Claude to pull information from the web for that date. Searching in Claude may take two to three minutes, so be patient and wait for Claude to respond. Depending on how fresh the AI news is, example, 10 minutes versus 10 hours, there may or may not be information on the web. If your first option out of the three does not turn any reputable or verified results in additional information, then move on to the next AI story that you found until you found one that returns enough information. Lastly, take that information from Claude and then use the connected Canva app and create a simple yet detailed three-page explainer on the topic. Then I'm just saying go through, do this step-by-step, don't stop, all that good stuff. I'm going to hit go on this. I'm going to very, very quickly click over it. Let's see if I can do it. We'll see if this works. I'm going to very quickly click over and hopefully, I don't lose everyone. Let's see what could go wrong. Again, I hate when people do this. Oh, look, I'm not using hands. But in this, I'm not using hands. Anything you see happening on my screen will hopefully be Codex. Right now, I just want people to be able to hopefully see the cursor. Hopefully, this works, right? So on my screen, I do have Chrome open. There you go. There's the look at that magical cursor. So Codex has its own cursor. You can see my cursor on the screen. So it opened a new tab. It went to Twitter or x.com. It's clicking the home. You'll see how my gosh, if you've ever used any computer use products, you can instantly see how fast this is. This is just lightning fast. So right now, even though I told it not to go on the 4U, all right, it went to the 4U. So now I'm going to have to, this is interesting, right? It's opening, right? I'm hoping it's not going to accidentally dox myself or anyone else. So far, so good. It actually opened my gmail.com in the browser, not my Gmail app, even though I told it to, but that I should have clarified probably multiple times. I do not go to gmail.com even though I did say use the gmail app and hit at on the app, right? When I ran this in 5.4, it didn't happen, but 5.3 Codex is much faster. All right, so I'm actually going to switch over here and kind of read what we have going on, because one of the things that I told it to do was to very much describe what it's kind of doing step by step, because what I found is if you don't do that, you only really get a sentence or so, and then it's hard to iterate and improve that prompt, because the ultimate goal here of what we're trying to do is we're trying to get a consistent result that gives us a good and verifiable output. The reason why is because I want to turn this into a schedule, something that I can say as an example, we'll run every single day at 7 a.m. So I wake up and a ton of work can be done, or this could be the first step in a series of schedules. So I want to make sure that I can get consistent outputs reliably. So I already saw one thing is even though I tagged the Gmail and told it to use the connected Gmail app, instead it went to gmail.com, which I did not want it to do. Anyways, let's see, it says, error running remote task, your input exceed, okay, yeah, fine. Let's try this again. Yay, live demos. All right, let's try it again. Here we go. So what I'm going to have to do here is copy this, so bear with me. And if you look at the left side, inside Codex, you see only one has an error sign. Obviously, it was the one that I did live. I did this three or four times beforehand and everything worked perfectly. So the demo daemons, fun. All right, let's do this one more time. So bear with me. I'm going to have to click the app button a couple of times. All right, so give me a second. So for this computer use, I'm going to have to click that and computer use. We'll see if I can do it. All right, and then I'm going to say Gmail app, and then I'm going to tell it this time. Then using my connected Gmail app, and I'm going to say, do not use gmail.com or the browser. You must use the connected app. All right, then it's going to say, find today's everyday lose letter. There we go. Let's make sure, sorry, y'all. Nothing more I love that's troubleshooting agentic AI live. All right, so all I'm doing is I'm just retyping these. So when I paste it in, it doesn't carry over the at mentioning or at tagging. So that's what I'm having to do. I don't know, live stream audience, tell me, tell me what are your plans for this afternoon, right? Let's keep everyone else entertained. What big AI project are you trying to knock out? All right, so let's make sure, I think I had one more. I think I have my Canva one here. There we go, the Canva app. All right. Let's see, let's hope this works now. Let's hope this works. All right. All right, so I'll just do that again. I don't know if mentioning the same app multiple times is eating into the context window. I'm not sure. I haven't tested that, but that's obviously something I'm going to try to do. All right. Let's hope this doesn't fail. All right. So same thing, I'm bringing back up my Chrome here. Hopefully the first step is it should open up a new tab or take over the current tab and go to Twitter or apps.com and then kind of browse around and look for some trending AI news stories. How did I know this was going to happen? All right. But hey, at least you get to see a second time. That really cool, right? So you can see my cursor moving on the right-hand side, and then you can see Codex's cursor moving on the left-hand side. So it's scrolling through Twitter, which sometimes there's relevant stuff, sometimes it's not. All right. So apparently it already got all of the topics that it needed. So this time it is correctly using the connected Gmail app. So I can see that when I'm looking at what's happening on my screen in the main Codex panel, I see that it properly called on the tools. One of the tools that it called on was, we're going fast. It correctly called on the Gmail app. But if you notice, it already opened up Claude. So now you look at the bottom, you can see my Claude app on my computer is open. I didn't have to do anything. It wasn't open when I started. So it found, let's see what it found. So talking about the Anthropic $100 billion compute, OpenAI's GPT-2 image drop. So it found some different things here. Let's see how it's actually using Claude right now. So all right. It looks like this is funny. Yeah, Claude just threw an error. As always, Claude's downtime is absolutely garbage. All right. So it did properly open up Claude. It gave it a prompt. It said, research the latest web coverage on OpenAI launching GPT Image 2 and ChatGPT. So it did what it should have done. It looked at Twitter. It saw what we didn't cover in-depth in the newsletter. It found one thing because by the time we set our newsletter out yesterday, it was right before GPT Image 2 was released. So it spotted that. Now, it's having Claude pull this information. So there we go. It's doing a good job. I don't want to hijack the work that my coworker is doing, right? So I can look over in my own monitor here. So now, it's looking at Claude. It gave it time to run. So it looks like, let's look. Yeah, it looks like Claude is still working, but it's almost done. I did tell it in the prompt, it's going to take like two to three minutes. So give it time, be patient. I can see the Codex cursor there kind of float around so I know it's still working. All right. So it says, Claude finished. It produced a usable source linked research brief on GPT Image 2. What changed? Who gets it? Tears, implications and uncertainty. Then it says, next, I'll generate a three-page Canva explainer directly from this material and finish with a concrete openable design link. So it is working correctly. There we go. So the cool thing to know about Codex, right? It obviously has way more capabilities than what you would use inside ChatGPT or the ChatGPT app, the desktop app. Because not only does it have MCPs, Model Contacts Protocols, so you can connect it to literally anything. You can connect it to almost literally anything and automate anything, right? When you think about Zapier or just how easy and how readily available MCPs are for literally anything, you can now start to automate anything. But Codex, remember, it has read write access for any folder that you give it access to. So it can access all of your local files. It can use your terminal, right? Which is a big thing because that really unlocks a lot of capabilities on what you accomplish while you're away from your computer. All right. And just like that, we're done. All right. So we did have to run this a second time. I don't know why I did this. Like I said, I did five tests and they all worked well, but such is such as life. All right. So here we go. So it says Canva output created. So it went through. I don't know why the previews, the generated previews aren't working, but it did leave a link. So let me go ahead and click this. The three page explainer. Or do I need to go to the edit link? Let's see. Bam. Look at that. So we have a decent looking presentation, right? This is probably something way better than I could have put together in two minutes. All right. So it says, let's see, three pages, just as I asked. There's a cover page, GPT image 2, the key features. There's the instant mode, the thinking mode, improved quality, enhanced outputs, multilingual support, and API uncertainty. And then for each of those six categories, it does have an image. So in the images are actually relevant and makes sense, which is fairly impressive, right? So I don't know if that's more, if I should be more impressed with Canva or Codex in this case. I don't even know which AI did the work, right? Who knows? But it's actually a really nice looking deck, right? It did it in 30 seconds. I could have had a template saved. I could have given it style considerations, things like that, right? Like the new Claude design, which is really impressive. But I didn't. I just open-ended, said, go do all this, put it together in a three-page deck, and then it has the business impact of generative images. So overall, it created a very usable deck that came from a very general prompt, right? But I wanted to start very wide. And if I did want to use this, I would obviously iterate and improve on this, right? But you just saw there, obviously, with a hiccup, how this can be an absolute game changer. Maybe one day I'll do side-by-sides of Codex versus Claude Code, Claude Cowork. But if it involves computer use, at least right now, I can't make you guys suffer through that. The computer use, and I've been saying this since it came out, it's obviously super impressive, or it was super impressive, but now it absolutely stinks. It is slow and ineffective. For people on the $20 a month plan, on Claude Cowork, if you're trying to do anything with computer use, like good luck, you're going to completely eat through your tokens by looking at it the wrong way. Your rate limits are going to be hit and you got to wait another five hours. Very impressive, right? So let me just very quickly recap what we did there. All right. Again, second time. But, and this is why I need non-technical people to get over the fact that this program is called Codex because I didn't write any code. I didn't see any code rendered on screen. I didn't have to wave a magic CSS, Ajax, NextJS, WAND, right? No, just did everything for me. So it used my actual computer. It connected to apps that I had already connected with my dynamic data. It opened and used programs on my actual computer that it didn't have a direct app for. So just figured all that out on the fly. Right? This is absolutely amazing. So I do want to quickly give everyone a very, very brief overview of the Codex app, right? I know we're already 40 minutes in, ran into a little hiccup, but that's fine. But if this looks familiar, if it's your first time, well, it's essentially ChatGPT, right? Which is, again, you can see why they only really need one app, especially once you bring in the browser capabilities. But here's essentially how it works, how Codex works. On the left-hand side, you have things like New Chat, Search, Plugins, Automations, and then you have Projects. Projects don't work the exact same as they would work in ChatGPT. They're a little bit different. So let me explain. So in this project, I have what's called Zapier test, right? So when you set up a project here, you can or you do need to have a folder that it is linked to. So what that means is each project needs a folder to output files, right? Or to create or to read files, right? So you can give it access to your downloads file, your documents. I probably wouldn't do that because, yes, mistakes can happen, especially if you're new. So I would probably try to sandbox this in a just a dummy folder as you're learning it. All right. So aside from that, that's the only big thing. And you don't really have to worry if you're a beginner about things like the repo, whether you're going to work locally, whether you're going to push it to GitHub. You don't even really have to worry about that unless you're trying to build a production app. But for non-technical people that are trying to get work done, this is going to look and feel very much like ChatGPT. And then there is this new feature down here just called chats. This wasn't here last week, right? But essentially that just allows you to quickly chat without giving it access or requiring access to a certain folder. All right. So aside from that, it is very simple to use, right? There is a search bar, there's a permissions thing. So there's default permissions, full access or custom permissions, and that's all. So when it opens different files, if it's using the terminal, all those different things. If you put the default permissions, it's going to ask you every step of the way, because a lot of those things may carry some risk, especially if you don't know what you're doing. Or if you just give it some wild commands, you might just be like, hey, I'm going to do this. Do you want to allow? Do you want to allow once or allow for this full session? Or you can do full control, kind of the YOLO mode. If you know what you're doing, if you have it sandboxed, right? If you're an entrepreneur, sold a entrepreneur, business owner, whatever, there's less risk to doing some of these things. So that's aside from that, it is very much just like using ChatGPT. You have these backslash commands. So you can fork something, you can bring in an MCP, you can access your memories, you can switch the model, personality, go into plan mode, see the status, check your rate limits, all those things. There's skills in the same way that again, it's crazy to me that there's no skills in a $200 ChatGPT Pro plan. This is why I don't use ChatGPT on the web much anymore. I use it for deep research. I use it for certain GPTs, but I'm kind of converting all of those over to skills anyways. So you have all of your skills there. And then if you click the at key, that's where you have your different plugins. You can upload files just like you can inside of ChatGPT. One difference you will see is there's something called a plan mode. So that essentially just breaks a complex task down into a plan. All right. And that's very helpful, because this thing can work for a lot of hours at a time. If you crank it up to GPT-54 on extra high, and you go through plan mode and you give it a huge refactor, mainly if you're doing coding or something like that, or I did have it go to every single page on my website and save all the transcripts, that took a couple of hours. But this thing can work for a couple of hours for you at a time. All right. Aside from that, I'm not going to go through all the settings. You all are smart. But two other things I did want to show you is plug-ins. In the upper left-hand corner, you have all these different plug-ins. Everything from computer use, that's technically a plug-in that you install. You have to change some accessibility settings, give it access, all those things. But there's Notion, Google Calendar, Teams, Outlook email, Figma, Outlook Calendar, Versel, SharePoint, Google Drive, Linear, Slack. A bunch of coding tools, Netlify, even though I'm focusing this on our non-technical audience. Canva, ReMotion for Video, Productivity Tools. I mean, you have Stripe, Box, Jam, Adio, CRM, Brex, ClickUp. A lot of just common tools, Firefly, Granola, HubSpots, monday.com, Pike Drive, SEMrush. You think, oh, this is for coders. I mean, what I just named off there, what I just rattled off, that might be half of your tech stack. That might be two-thirds of your tech stack. All of a sudden, you're like, wait, I could probably automate a good chunk of what I'm doing every single day. You might be saying, wow, why would I ever use chatgpt.com again? Again, if you have access and availability to use Codex. All right, and then last but not least, automation. So I'm actually changing a lot of my automations since the update from last week, so I don't have a lot in here. But in natural language, you can just create an automation. So you can set it to a certain work tree. So that's something from GitHub, or you can just run it locally or in the chat. You can select a project, so you can set up an automation to run in a certain project, and then you can schedule it for when. So daily, weekdays, weekly, custom, etc. Then you can select the model that you want that scheduled run to go on. So for something, if there's a lot of computer use, maybe you want to do one of the Spark models, which is a little bit faster, and then you can set the reasoning level as well. Then there's also templates for kind of automations. So that's it. Kind of bumpy, slightly bumpy live demo, but it is what it is. All right. So let's quickly wrap this one up, and I want to give you guys the blunt truth. I don't care if you're a non-technical user, you need to use Codex now. It is that good, and here's how good it is. I don't even see myself using ChatGPT on the web much in the future, at least not right now, right? I mean, you have the, you know, oh, I love the new GPT images too, right? It's built in, right? Presumably, you know, Atlas at some point, some of those capabilities will be built in. So the only thing I'm really using ChatGPT on the web for is deep research, and there's certain things. I like the chain of thought on GPT 54 Pro better in chatgpt.com than it is in Codex. It's a little more limited and I really like to learn and improve, you know, kind of my prompting skills and how I piece together different workflows. So I do like reading the summarized chain of thought a lot better, both in deep research and in GPT 54. And then, like I said, maybe for some GPTs, the majority of my day-to-day work is no longer happening in ChatGPT, it's happening in Codex, right? So I also will say this, hot take right now, Claude is extremely fragmented, right? Because you might be looking at these things and you could say, oh, I can do most of these things in Claude code or Claude co-work, right? Aside from the computer use, yeah, you probably can, but it is absolutely fragmented. It is siloed. And that's the thing that I couldn't really showcase live. But the importance of persistent memory under a unified, agentic context is huge, right? Because if you're using the Claude desktop app, aside from the computer use being absolutely terrible, Claude code doesn't know what Claude co-work is doing. Claude co-work doesn't know what Claude chat is doing. Those are three different areas. So at times, even for myself, I'm a power Claude user. I do love Claude. I love Claude desktop. But even at times, I find myself saying like, which one should I use? Sometimes I configure an MCP server on the web, and it doesn't show up on the desktop. Sometimes I configure it inside Claude code, but then it doesn't show up on Claude co-work, and I can't even reconfigure it because it says it already exists, but I can't use it, right? So there's all these things that the chat co-work Claude code layout of the desktop. It's actually very annoying for me, right? I do think Anthropic has hinted that in the future, it will be a unified system, but I think this is one of those instances where Claude code or Claude in general by being first to really dominate this kind of agentic desktop experience might have actually set itself back. Because I don't know if this was OpenAI's plan, right? But I can assume, right, they, you know, in February, they put out their version of Codex and, you know, Anthropic's been shipping like crazy and they're probably seeing and understanding what people are complaining about. They're like, wait, we can just create a better version of this. Sometimes it might be better to be, you know, the second, right? You know, obviously, Codex has been out for a year, but the desktop app has only been out since February. So right now, they just don't work well together. And Codex, like I said, it's probably a top three AI tool or release, at least for me ever over the past three years. And I don't say that lightly. Probably Notebook LM in the ChatGPT Plus, the first paid version of ChatGPT would probably be those two other moments. This is huge. So I'm wrapping it with this. Use Codex. Now, whether you are a heavy ChatGPT user, whether you are a heavy Claude Code, Claude Cowork fan, you need to be using it because there is a period, I don't know how long that period is going to be, but it is such a cheat code to understand the basics. I gave you the five steps, go follow them. So once you give it some time, once you refine and perfect some of your workflows and schedule those, again, maybe everyone comes out with this in a week. I don't know, but at least today, if you do these things, take some time, set it up, improve it, it is an unfair advantage. Go take advantage of it. That's it. If you do want access to whatever our future cohort or course is, the co-working court, make sure to repost this on LinkedIn. Then if you haven't already, please go to youreverydayai.com, sign up for the free daily newsletter. Thanks for tuning in. We'll see you back tomorrow and every day for more Everyday AI. Thanks, y'all.

Speaker 1:
[53:25] 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 youreverydayai.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.