Ep 887: ChatGPT Work: What’s New, Who It’s For and How to Use It (Replay)

There's a viral new product from OpenAI that's melting servers and surging in popularity right now. It's called ChatGPT Work, and it's new on both desktop and the web, though they're kind of different products there. So the company launched ChatGPT Work last week in conjunction with their new GPT-5.6 model, and it seems the current AI conversation among non-technical users is legit swarming around this new product push. So much so that OpenAI says they've literally added 2 million new users over the past 2 days, which for a company that's been around commercially for like 3 and a half years, launching a new product line and adding 2 million new users is kind of unheard of. So what is it? Well, ChatGPT Work is technically not new at all because it's just Codex for beginners, which I think is actually a great thing, though there are still some minute differences. So what's the point of this new ChatGPT Work on desktop? How is it different than ChatGPT on the web? And should you just use Codex instead? Well, we're going to be answering those questions and a lot more on today's Everyday AI, but Let's put AI to work on Wednesdays. That's what we do. And here is the big picture of what's going on. OpenAI, well, they, with the new ChatGPT Work, they just kept Codex's core agent capabilities and changed the doorway, right? So it's easier for people to enter in. It's not, they don't see the word code and they're like, oh my gosh, this is scary. It's not for me. ChatGPT for Work. Wait, I work. I know ChatGPT, right? It just Made sense. So Work kind of removes that developer-first framing that Codex had, and it emphasizes finished business deliverables. And the response shows that non-coders were maybe just waiting for permission to use it. So on today's show, here's what you're going to learn. You're going to know what OpenAI announced and why millions of people immediately paid attention. You're going to know what changed from Codex, what stayed, and why the backlash online is kind of missing the point. And you're gonna know how desktop work and web ChatGPT work. Yeah, they're different. So there's a ChatGPT work desktop and a ChatGPT work on the web. You're gonna know how they differ, how they're same, and which one you should probably use. All right, let's get into it. Welcome to Everyday AI. My name's Jordan Wilson, and well, we do this every day. Uh, this is your daily livestream podcast and free daily newsletter helping everyday business leaders like you and me Not just keep up with AI, but how to use all those tools to get ahead, to grow your company and your career. So I always say like your Everyday AI, FYI, 'cause you all tell me what to do. You know, I put a poll in the newsletter at least once or twice a week. I let you all change the shows. So for people out there maybe complaining like, oh, you're covering too much OpenAI ChatGPT. Well, you voted for it, right? If I'm being honest, there was 2 products that I wanted to cover on today's show. I wanted to cover the new GPT Live voice mode on the phone, and I wanted to, cover, uh, Claude Design. But you all wanted ChatGPT Work, which is great because I technically have been using Codex for like 12 hours a day for the last 5 months anyways. Uh, so let's get into it. And if you haven't already, please go to our website at youreverydayai.com. We're going to be recapping the highlights from today's show and all the other AI news you need to know to stay ahead. So ChatGPT Work, simple name. A lot of confusion over the rollout, and that's okay. And let me actually just address that first before we go live and we go hands-on. There was a lot of backlash and people were confused because when you went to update Codex, as an example, it looked like Codex was broken and then it started downloading something called ChatGPT and people were like, what the heck is going on? Right? So there was a lot of drama in the first like 24 hours. People were confused and they're like, you're taking our Codex away. Right? But As you'll see on my computer right now, even though it is still buggy, it kind of changes, right? I still have the Codex icon, um, even though it says ChatGPT when I hover over it. So this is the super app, right? This is bringing, uh, kind of the, the best of Codex, the best of ChatGPT, and the best of their Atlas browser all into one package, but giving it technically 2 different doors, right? You can go in the front door or you can go in the dev door. Uh, you know, off to the side. All right, so let's get this thing started live. All right, so livestream audience, help me out. Let me know if you can see my screen. Uh, and we're doing this one a little bit different. All right, uh, maybe you can see I'm, I'm traveling, I'm in a hotel room. Uh, so actually prep for this one was kind of interesting. I'll explain it, but this is all very meta 'cause I used Codex to build this. You know, if you saw my slides, they're a little different. I actually just did a Codex/ChatGPT work presentation with the Affirm team. So thanks to Affirm and Nanika for inviting me out to do that. And, you know, got to work alongside the OpenAI team as well. But one thing that I really wanted to call out is there's a lot of excitement, right? A lot of times when I'm doing these trainings and I'm like encouraging people, Uh, you know, bust out their laptops. You know, maybe it's like 50/50. In my second session, I think literally every single person had their laptop out and they were working on ChatGPT Work, which is pretty cool to see. All right, anyways, our livestream audience will be able to see this. Podcast audience, you can always see the video version at youreverydayai.com. So what I have on my screen here, well, it's ChatGPT Work, right? It looks kind of like ChatGPT if you've never used ChatGPT Work, it should look familiar-ish. All right, so here's, here's all the, the hoopla, right? In the upper left-hand corner, you can actually just change— there's a little dropdown, super simple— and you can change to Codex. All right, and you'll see nothing really changed on my screen. I'll get to the differences later, but if you're confused, or if you open it once and you're like, wait, why does it say Codex Don't worry. It is literally the exact same. You can change midway through an action or a task or a prompt, and nothing happens. Right? All your tasks stay the same. Your your projects, everything stays the same. It just essentially changes the view. All right. So let's get that part out of the way first. And now let's get a little meta, shall we? All right. So here's what we're gonna do. We're gonna have a little fun because I'm gonna technically show off one of the features of ChatGPT. Work by using my phone. Okay? Because one of the great features, if you have the ChatGPT app, which I do right here, well, there's a remote setting. So this is actually how I was building my presentation, you know, putting some finishing touches on the plane, right? I was just using remote mode on, on my phone, controlling my home computer, which is really cool. So yeah, you can hook up if you have multiple computers that you use on the same ChatGPT account like I do. I can go in and check on my computer at any time, which that's one of the big changes, the big new capability unlocks that we'll see. It's a little bit different, right? Depending on what your organization allows you to do or not. But if you do have that ability, this to me is so big, right? It's allowed me to literally go out and touch grass. A little bit more because at any time I can access any file on my computer. I can kick off any process, any automation, just by talking. So let's see how this goes. Hopefully it goes okay. So all I'm going to do is I'm going to open a new task and I'm choosing my MacBook Pro, which is the, the one that I'm sharing here, obviously. And I'm just gonna, I'm just gonna yap. A little bit. All right, so here we go, and let's go ahead and— all right, I'm just going to dictate this on my phone. For context, right, I'm using the new GPT-5.6 SOL model. I'm using low just so hopefully it'll go a little bit faster, and I am using fast mode, which consumes more credits. So here we go. Look at the Beehive MCP and look up my newsletter from July 14th. Find the 5 stories that drove the, drove the highest number of clicks or a much higher number of clicks relative to where they were featured in the newsletter. Exclude any domains from youreverydayai.com or spotify.com, as I don't really care about those. Use the context you know about me and Everyday AI and what our audience cares about. And then after all of that, write a short 5-bullet-point recap describing Those 5 stories. So 5 stories, 5 bullet points each describing what's new, what happened, the nuance in the background required to understand that as well, as that's pretty important. Also, which is very crucial, I need you to find the folder on my computer called Final Everyday AI Transcripts. So you can use computer use and look in the downloads folder and make sure you go through the last, like, I don't know, 2 months of transcripts to better understand the background context, trends, latest happenings to better frame those 5 AI news stories. So treat those transcripts as like your first source of truth, but search and verify any missing gaps from multiple reputable sources from 2026 only. Then use the sites feature and turn this into a mobile optimized site. Nothing fancy, just Easy to read and understand, and then publish that site. Uh, when you're done with all that, use computer use again and use messages on my computer to send Jordan a message with the link to that site. And I do approve, uh, you sending this message ahead of time, so please send that message. All right, so I sent that. It's already done dictating. Pretty cool. See it pop up there on my screen on my iPhone, and then if I pop open this right here, there we go. So it dictated my my query there straight from my phone, put it all into ChatGPT Work, and ChatGPT Work is getting to work. Work. All right, so I'm gonna first kind of, um, I'm actually not sure how I'm gonna do this. Um, okay, I will kind of explain a little bit more on the interface, what's going on, and then hopefully during that time I can tell you exactly how ChatGPT work is, well, working. So, uh, right now, yeah, it's not technically doing anything on my phone, though I can see it, uh, right So a lot of people I think were confused by the remote because they're like, wait, can it grab things off my iPhone? No, all this is is a remote. So previously this was called Codex on the iPhone. That's also confusing. I think it's much better to just be called remote because that's all it is. And I would love, by the way, if, and I know this sounds like counterintuitive, but I would love that remote feature to be on chatgpt.com for people that have multiple machines. I think it just makes sense, right? Especially for me, like when I'm traveling like now, I would love to be able to just 'Cause I'm always on my computer to be able to grab something via remote on there versus having to dictate it. But it is what it is. Okay, cool. I hope that it was gonna do this. So you'll see in the upper right-hand corner, there's this new picture-in-picture feature of computer mode. All right. Uh, or sorry, of computer use. So right now it is using computer use. Um, and you can kind of see it. I'm not gonna open that little tab 'cause it's gonna show like literally everything in my downloads folder. Um, but you can see that little tab up there in the right-hand corner. Which shows me that computer use is working, which is really cool. All right, the computer use overall is another area that ChatGPT really highlighted last week along with the, obviously the new ChatGPT Work and the new GPT-5.6 models was the improved computer use. And I like this new little feature. So it really just increases the transparency. So let me just kind of show you the difference between what you would see in ChatGPT Work versus ChatGPT Codex. And well, it's not a lot. So when I click over to Codex now, All you're gonna— okay, actually right now you're not gonna see anything until it starts to report back on the findings. But all it really does is simplify the amount of information that it shows you, right? It's not gonna show you all the code that it wrote to get this task done because that's all that's happening in the background, right? And you can see it's obviously using my computer, but it's also running code to do all those things. So essentially Codex, you know, one of the people that people are apparently Scared of it, even though they're, if you're using it for non-technical work, you know, people just having the word code in there and they say, well, I'm not a coder, right? It's almost like showing up to a party you don't think you're invited to, but you definitely are because there's millions of people just like you on the inside, like me, who are starting to automate all their busywork. All right, so we're going to let that cook, see how it does. Hopefully it does well, right? But let's get back to the details, talking about what's new, all that good stuff. So yeah, they added 2 million users over the last 2 days, which is pretty bonkers. And that's on the ChatGPT side and on the Codex side. So yeah, kind of same thing, but having 2 million people bring agentic AI to their desktop over the course of 48 hours is a huge signal of where this appetite is, right? So to me, you all, if you listen to the show, you know, I've kind of been this crazy guy for the past couple of months being like, if you're not using Codex, you are literally missing out. And I can't count literally the number of conversations that I had with people who just couldn't get over the word, right? So I do understand some of the backlash of the heavy Codex users because they're like, Well, now sometimes it says ChatGPT. It's like, who cares, right? Like, you can still say you use Codex because it's still there, right? All the getting the diffs, getting, you know, now there's a new pull request feature. You can still get all the technical stuff, but it's not just a better version of ChatGPT. All right, the biggest difference here is it can use your entire computer. You know, it has its own built-in in. ChatGPT Work has its own built-in browser, uh, which is really, really cool. Maybe I'll show that here at the end when we check in on the, uh, on the results. But it works just with you, right? That's the thing. Kind of what I said in my presentation is I said I think the chatbot era, the early ChatGPT, was like you're handholding an intern, um, kind of the early agentic era, um, which is, you know, the When we had the GPT-5, right? The early Claudes before Claude Code on the desktop, right? That was just like having a smart coworker, right? Working along with you. Now that we have agentic AI and the big step here is, well, I can have hundreds of these things going on at once, right? I can have dozens of agents using my computer, right? It's not just one thing running. I can have multiple tasks scheduled. I can just command them using my voice. And, you know, these things are running around the clock. So it is something that now that we've entered what I say is this third phase, you know, so we've gone from handholding an intern to collaborating with a colleague to now you are directing a team of experts. So I do like how OpenAI timed this new release with ChatGPT Work, bringing it to more non-technical people. But also at the same time, coupling it with this new model, GPT-5.6 Sol is an absolute beast. You know, you're getting Fable 5 level actually above, I would say on most, if you look at a lot of different benchmarks. So you're getting Fable 5 level, but at like a third or a fourth of the cost. So, you know, that's the big difference there. So like I said, Codex is still there, and there has been some, some criticism, which I understand and do agree with, um, because you technically now have a lot of different versions of ChatGPT work, right? So you have ChatGPT work on mobile— or sorry, not on mobile, um, on the web, right? But the web ChatGPT work doesn't sync with desktop ChatGPT Work. And you also have ChatGPT Work on mobile, and ChatGPT Work on mobile syncs with Work on the web. So there are some things where it's like a little confusing, but then also chats from ChatGPT on the web do sync with ChatGPT Work and Codex. So those, at least you do have some level of syncing. Uh, you know, it's not like Claude Code where If you open up one thread, it has no clue anything what's going on, right? That's another thing I love about Codex or ChatGPT work, which is good call to call out any thread inside any new chat that you create, uh, any new task. Um, you can say, ask it about literally anything you've ever done in any other chat or task, uh, which is a huge unlock for people that sometimes get unorganized. Or for me, right? I'm always firing off a lot of these from my phone, which leads me to be a little less organized than I should, uh, but to be able to go and ask any thread to go find any information, and you can also have it clean itself up. You can say, go rename all these and tell me what's on my to-do. So from any, uh, chat or from any thread, you can control anything inside of ChatGPT Work. But it is a little bit more confusing now. Uh, I'm sure OpenAI is going to work on, uh, you know, communicating this and probably unifying the experience a little bit more. Uh, but essentially now you have chat, Work Codex, um, and Work ChatGPT Work on the desktop and ChatGPT Work on the web and mobile. All right, so I've talked a little bit about what makes it more powerful, but let me just quickly, uh, recap some of the actual features and functionality of what made Codex special, which now you have inside of ChatGPT Work. Uh, well, It can access anything you give it access to, right? You can, I wouldn't necessarily recommend this, but in theory, if you know what you're doing, if you put proper guardrails on, you can give it access to your entire computer and it can read, it can write. You know, so a lot of times I have agents off doing these mundane tasks for me and they'll just save all of that information to my computer and also, you know, back it up on Google Drive. And then obviously, just like ChatGPT, it can create end artifacts as well. So I'll say this, you know, between being able to read and write any files and folders, the level of intelligence, being able to create any artifacts, but then being able to also spin up subagents by itself, right? You can literally just say inside ChatGPT Work, if you know it's going to be a huge task and you give it a ton of context, just say, break this up with 5 subagents, right? Essentially, that makes your Task complete 5 times faster. You know, you're technically paying, whether you're paying, you know, the token price or you're paying in usage credits. You know, it's gonna eat into your usage a little bit quicker, but sometimes you just gotta get stuff done. So you can just divide the work up into sub-agents and get things done much quicker. So if you've ever worked on a big project in, you know, ChatGPT on the web or Claude on the web or Gemini on the web or Copilot, and it's just one kind of chat thread and you're like, Hey, I'm getting a good result here because I'm using this very powerful agentic model, but it's taking forever. You're gonna love working with ChatGPT Work because not only can you direct it to do subagents, but a lot of times it's just gonna do that on its own because it's super smart. It's gonna say, hey, I should probably, you know, have a research agent, a design agent, and then a QA agent to make sure that everything that you said and everything these other agents do is correct, right? So you almost have this like managerial agent hierarchy that they set up on their own to make the outputs that much better. So let's just quickly talk about the harness, okay? Because ultimately, the harness here is really what sets it apart from using anything on the web. And that's what brings together, you know, your actual model, the context that you throw at it, and it handles a ton of context. And anytime it runs out of the context window, I think Codex, ChatGPT Work by far are the leader in auto-compaction. Not even close, right? So when you do go over the context window, I don't know what kind of wizardry the OpenAI devs do, but it'll essentially crunch up and just send only the important takeaways and keep it in the context window. So working with the model, the context, the tools, all of the rules that you set, there's agent MD files. Don't worry if that sounds daunting. You just say like, Yo, uh, chat, make me, uh, you know, an agent's MD file, uh, to govern this project according to what we talked about. All that is, is it's a text file that tells the agent what to do or what not to do, right? So by default, there's an agent MD file that oversees all of your work inside of ChatGPT Work on the desktop. Just like, you know, you can set custom instructions in ChatGPT on the web, but then in an individual project, uh, project. You can set custom instructions there. The same exact thing happens on ChatGPT Work on the desktop, but those are just called AgentMD files, and that's just a set of instructions. So, uh, goals are extremely valuable as well. So goal, you can just give it a goal and it will literally work until it gets there. That goal may take 5 minutes, may take 5 hours. It may literally take multiple days. Uh, again, keeping, keep in mind your usage, but if you give it enough context, enough of your decision-making prowess, uh, And if you're ambitious enough, it will actually go out and work for a long time. So talking about some of the, these are slash commands. So if you are maybe listening and opening your laptop, which I always encourage you guys to do on these Wednesday shows when we put AI to work and we do these kind of live demos, just hit the slash command and you're gonna see all these different features and functionality. Plan mode is great. That's essentially the, the basics of chain of thought prompt engineering that you go through side by side. With ChatGPT work to give it better context. You can steer it along the way. So that's another great thing that most ChatGPT users are gonna really like because the casual user has never had that power. You've always had that power if you were on the pro plan, not always over the past 6 months you've been able to do that. But being able to steer or cue is big, right? So you saw me, I just dictated that very long message. What happens if I made a mistake, right? Normally, you either have to hit stop and start over, or you literally have to wait, which could be 10 minutes, could be 10 hours, and then you're like, okay, I gotta follow up now, right? So steering and queuing, you can literally send 100 messages at once, and you can steer. So to steer means, oh, you send it to the agent right away while it's still thinking, and it will ingest your query and make any, you know, modifications. Queue just means, okay, once it's done, that next message gets sent automatically. So I always queue up a ton overnight aside from automations and things like that. So loops can just repeat processes, kind of like a heartbeat. If you've used the OpenClaws of the world, you know, you can just have something go every 30 minutes as like a heartbeat, or you can just literally keep looping and iterating. I think if you're in software development, that makes a little bit more sense. You know, when you have all these PRs that you have to fix, you can just literally just keep looping those. And then we talked about those subagents as well. So some of the highlight features, again, that really separate this from ChatGPT on the web. So being able to use the browser and the computer, apps and plugins are available on both, skills and automations are available on both, and then the remote as well. All right. So that is a wrap. But like any good cook, let's check in. So let's first see, did I get a text message? From myself. Okay, so weird. It looks like the text message— let me just go ahead and share my screen here, see what happened. I did not get a text, right? That, that's why we do live demos. Sometimes live demos work perfectly, sometimes they don't. All right, so this did work for about 6 minutes and 50 seconds. And now maybe let me just go ahead and switch. Maybe we can see the difference here, uh, between ChatGPT and Codex, right? Okay, so you guys just saw it. It's like nothing. All right, so all it did is in this case, right, it took away, uh, what files it changed, right? So in Codex, I can see, oh, it, you know, it edited these 7 files and I can review the changes and, you know, look at the JSON and, you know, all these other files. So, you know, if you do get intimidated by code, well, my response in ChatGPT work Had zero code on there. All right, I can still click the, uh, to look at the chain of thought and see exactly what it did, which I always encourage you to do. So I should have explained this when I gave you that long prompt. Why the heck did I do this, right? That process right there, I don't know why I didn't build this skill like 6 months ago or longer once the Beehive MCP came out. So Beehive is the email newsletter provider that we have. And one thing I'm always wondering is like, what do you all care about, right? What What are you clicking on? What do you read? And then I want to know more about what you guys want to know, right? I want to learn along with you so I can, when I'm interviewing people or bringing you a new show, I know your pain points, right? So that's one of the reasons why I always try to personalize takeaways. So hopefully when I'm talking to you all, it's relevant, right? So what's weird is says, okay, before I send the messages note, please confirm the final action. Send Jordan. I pulled, yeah, okay. So what's weird is, You know, send the message. I'd say that it usually works like 80% of the time, and that's why I even gave it the pre-approval, 'cause I've done this so much. I should have invoked one of my skills I have. It just overrides that little weird message. I don't know why it always asks you to double confirm, you know, sending text messages, which is fine. Okay, so now it's doing that, but let's go ahead and open up what it created. So this is called a site. So I did do a dedicated show on Codex Sites, but it's essentially kind of like a cross between like a lovable. So it's literally like you can just build and vibe code apps for yourself, for your team, uh, interactive dashboards or just, you know, interactive slides, little informational websites to share with your team. Uh, they are private by default, but if you are on a team plan, you can share them with your team or you can make it public. All right, so let's go ahead and look. I didn't use any skills for design or anything like that because I wanted it to be, uh, kind of quick. So let me just go ahead and, uh, let's just go ahead and open this. All right, and I will share that screen. I got a lot of screens going on here, so let's remove this and okay. Give me a second, y'all. I got so many screens going. All right, let me share, share this screen. So here we go. Pretty simple, but like it. It's simple, right? I'm gonna look at it on my phone. But here's the top stories that people clicked on, and then a personalized, you know, look through our readers, right? It says for Everyday AI readers, you know, this enforces, reinforces a recurring theme, right? All this stuff, because I have a lot of information about who our audience is and what they ultimately care So, you know, number 1, Satya Nadella says enterprises may be giving AI vendors their real competitive advantage. Number 2, Grok Build reportedly uploaded entire code repos, not just the file it needed. Number 3, Meta workers say AI-assisted layoff scoring penalized people on protected leave. Number 4, the DeepMind story about Demis wanting a US-led watchdog. And number 5, Perplexity turned computer into a persistent agent workspace. So what's great there is I don't have to go in and see, because I'm always wondering, I'm like, we put between our AI news stories and our little fresh finds, right, which are just these kind of bullet points, we probably cover like 15 to 20 different things, right? So this is helpful to me to know what the heck you all care about. So, uh, that's it. That's a wrap. That is the new, but not technically new, but still yes, kind of new ChatGPT work. So if you've been wanting more power, or if you've been agent curious and you're like, how does this work? Right? You have a ChatGPT account and you've been enjoying some of the maybe more agentic features, right? Maybe you use plugins and you use, uh, you know, skills on the web and you use these apps, but you maybe want a little bit more. The desktop is where it's at, right? I already mentioned, you know, computer use, browser use are some of my most used skills. Um, but then just being able to schedule these things and having these things run around the clock for you, update kind of your working knowledge, uh, and, and you being able to go in there and just direct them to get more work done, more manual knowledge work, by the way, it is a big unlock. So I hope you found this episode helpful. And hey, do we like the Codex-created slides? Right? I'm, I'm 50/50. It's kind of fun to create these, uh, slides in Codex. Normally, I don't know why, I actually used to really enjoy, you know, copying and pasting all of this information over into Canva. But today I'm like, I had started a bunch of research in Codex anyways for my presentation. So I'm like, let's personalize this and be super meta and do a Codex presentation in Codex about Codex. But I hope it was helpful either way. If you haven't already, please do me a favor, subscribe to the podcast. Uh, then go to youreverydayai.com. Make sure to 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.