OpenAI Codex Can Now Use Your Computer! (Full Breakdown + Testing)

So this week in AI, we see the most important product category for most consumers develop. Namely, OpenAI is making consistent moves on their professional agent for non-technical knowledge workers product. Basically, AI agent for the 80%. And their product named Codex has been continuously trimmed and adjusted so that it's more user-friendly and more people can use it. So for example, if I want to set up an automation where my weekly work brief runs and it pulls in emails, Slack messages, calendar invites, all of that, I could do it. And if you're familiar with this, you might be saying, yeah, this is OpenAI's version of Claude Cowork. And you'd be right. But there are some differences. And we'll be looking at that today in this week's episode of AI News You Can Use, the show where we look at all the AI releases and I show you the ones that I think matter most to you. And as mentioned, today's focus is really going to be on Codex, because there's been a bunch of updates. It's not just that GPT-5.5 came out last week, which is very focused on agentic tasks. And a more goal-driven prompting where you tell it what good looks like and it works backwards from that. But also they've made several changes to the interface and onboarding that tune this whole experience more for a non-technical consumer rather than a developer. And if you're not familiar with this, in short, this is ChatGPT with access to your computer with this idea that it should just run by itself and do work for you instead of with you. And I'm here to point out the differences to Cloud Cowork, not that this is just another product in a category that already exists. And let me tell you, the main difference is this is really good at using the browser. It basically is a mouse that works. That's the biggest difference that I've been able to pick up on. The team has also been running a bunch of tests, analyzing transcripts, and doing several tasks. I want to kind of highlight the most interesting one to me. And this was something we could not get to work inside of Claude Co-Work. It was a task where it was supposed to find a video file that was on your desktop. And basically it has to open the video files, look at the frames, analyze them, and then do that for all the videos. Claude Co-Work failed at this. Codex Actually nailed it. It's a thorough little agent, that's for sure. And as we talk about things that it does better, it has an image generation tool just built in. No need to connect an external MCP to generate images. So here it created this poster after I asked it to create a Google Form. And as we talk about creating a Google Form, I just want to show you this live. Open the browser and create a Google Form to sign up for a basketball game. Its ability to remote control the browser is actually impressive. Just watching it is kind of amazing. Now, while it does that, I'm quickly going to show you what you need to do for this to work on your machine too. Well, first of all, you need the desktop app that you can download here. And once you log in, it goes to work. And check this out, you have the little cursor here. The competitor products don't show you how the cursor moves around. And what caught my attention is the speed. I'm not exactly sure if this is some of the recent Codex updates or if this is GPT-5.5, but look at that. It's moving around at a speed that, Well, maybe somebody would have if they're a bit slower with computers. But the great thing is you don't have to be following it here, right? You can go into another chat and kind of check out what has been happening here. And ooh, what is this about? Ah yeah, the draft slide deck for an upcoming customer onboarding review that it generated. All right, so let me just open that in PowerPoint. All right, this is very similar to ChatGPT. Well, I do find that the tasks in here are a bit more thorough, but And here's my first tip. If you're looking for use cases of this thing, I found this article published by OpenAI where actually this and 9 other prompts were recommended. These are different things that OpenAI thinks consumers would be doing with it. Look at this, top 10 use cases for Codex at work. First one and the biggest one is a chief of staff, which basically reviews your messages, your calendar, and then basically tracks your action items. For this, you do have to give it access to, well, absolutely everything, which I I don't know, for me, I have not done that yet. If you wanted to do it, you could just head on over to plugins here. In Claude, this would be called connectors. It's the same thing. And you could connect up all these different applications into here. What we're doing in the example that is happening in the background is browser use, right? That's where it's creating the Google Form. But it also is computer use where it can go outside of Codex and remote control apps on your computer and browser, which might be interesting if you have something you do on your computer. Again, its ability to kind of control the mouse and navigate is really spectacular. On the benchmarks, it's slightly above the Opus models. But honestly, in practice, it just seems like this functionality concretely is just really solid. Whereas in Claude, it's pretty good. But I don't like doing those tasks. It's just they're not reliable enough. Now, as you see it build out this form, I want to kind of tell you about what changed, because they're really trying to make this shift from developer-focused tool to consumer-friendly tool. But man, I don't know, if you look at the settings, Just check this out. MCP servers, Git environments, work trees. These are all developer terms. This is built for devs. They're just trying to kind of silently and slowly make this shift from dev-focused to consumer-focused. Whereas Cowork is truly consumer-focused. Also in Cowork, I kind of just love this interface where you can go from cloud to Cowork in the press of a button. It's just all there in one made for consumers. Whereas this is a standalone thing and the menus are dev-focused. So yeah, there is our little questionnaire that it just made, no problem. So here would be my recommendation. If you've never tried one of these, and you have a ChatGPT account, well, you can just download this and log in with ChatGPT and start playing a little bit, get a feeling for this category of professional agents for non-technical users. It's going to be a huge deal moving forward, but we're just getting started. And as you work with it throughout now, it has way more prompt recommendations where it tells you what you could try. So it's kind of fun to just try, I think. It's never been easier. And again, you just need a paid ChatGPT account and then you can use it. Just don't forget to install either computer use or browser use to take advantage of some of those abilities. And if you're still wondering what to try, if those 10 use cases from the article, like the chief of staff or something like drafting slide decks, which we also looked at, isn't really your jam. Well, if you're in business, then I think this one, Workflow Audit and Automation Spec, this is fantastic. It basically helps you draft SOPs with an automation spec next to them so you can, well, map a process and see how that process could potentially be automated away. You could transfer data into Excel sheets, you could build forms, you could fill out forms, you could do all the ChatGPT stuff, including building dashboards that are a bit more sophisticated than what you would usually do in ChatGPT. You could find video files, and then if you have a video file, you could easily turn them into GIFs. You could create a transcript vault on your computer and let it work with that. And then if you find something useful, you could set an automation where it just runs that regularly. But outside of Codex, there's a lot happening in this category. And I just want to highlight a few things from this week. First of all, we got a Google Workspace MCP server. So in other words, it's a custom connector for your Workspace account, including Gmail, Drive, Calendar, Chat. You could plug any of those into, well, either Codex or Cloud Code, Cloud Cowork. That's another fun thing to try with this. In the same vein, Cloud Code and Cowork added a bunch of these connectors to Spotify, Instacart, Booking, and more. They're just expanding the library of things that you can plug into these apps. And everybody's just trying to make the best consumer agent, as you can see. Oh, and talking about connectors in Cloud, actually, they made a big move with Blender, Adobe, Autodesk, Ableton, all of the big creative apps basically now have connectors that are, well, not very smart. I mean, you connect Photoshop in and you would have the expectation of, oh, now it's going to Photoshop stuff for me. But really, it just kind of does the first, second step, and then you need to open Photoshop and go there. But it's happening. They're trying to connect all the apps into one big thing. And finally, a few significant other stories that I just quickly want to point out is one, we got a new Deepseek model. If you're not aware, that's China's thinking model competitor, completely open source. And it's so damn good. It's close to the state-of-the-art models. And while the big one is absolutely unusable on any sort of consumer hardware, it does come with a 1 million context window. And it's the new state-of-the-art open source model. And finally, this might be a bit random, but Google is integrating, or I guess testing, an AI chatbot search in YouTube. So instead of searching at the top, you kind of talk to the videos and you can follow up and chat with the videos. It's weird. Isn't that like against the point of YouTube? I don't even know. Kind of taking all of this in and seeing all of these developments just to me makes it kind of obvious where we're heading. They're trying to unify the internet into one experience. I think eventually it's not going to be like laptop, smartphone with 100 apps. It's more going to be like, hey, you talk to your agent, whatever interface that comes through, and then it reaches out and gets info from YouTube and uses your Photoshop application to get things done. And I don't know, creates Excel sheets and communicates with your colleagues and handles all the action item and deadline tracking. Everything I showed you in this video that right now you kind of have to do manually and do together, or maybe you set up the workflow together with the AI, will be more and more automated, which will, well, free us up to, I don't know, drink more water, go for walks. But yeah, that's really everything I have for this week. I hope this was helpful. If you want a quick overview of all the stories that were on the docket for this week, well, these are all the stories I was kind of picking from. But we just talked about the most significant ones. I figured actually showing the screen at the end of the video might be kind of a nice touch for people who do want to see other things that happened. And with that being said, my name is Igor Pogany, and I hope you have a wonderful week.

AI news you can use. Game plan power for the work you choose. AI news you can use. Hit that play, let it light your fuse. AI news you can use. Weekly upgrade, tighten up your moves. AI news you can use. Welcome to the show, now here's what's new.