Ep 781: OpenAI’s Codex For Beginners Pt 2: Advanced Agentic Strategies broken down for Newbies

We all wanted some futuristic AI that could be so advanced it could write entire programs for us and complete our big projects and triage our daily grunt work and run around the clock with or without our input. Oh, and you should be able to talk to it from your phone and control the whole thing even when you're not in front of your computer. Well, guess what?

That holy grail of AI day, it's here, and 99% of people are sleeping on it. It exists in OpenAI's codecs. And I've been doing this everyday AI thing for more than three years. And I'm not exaggerating when I tell you that individuals in companies that are going all in on codecs right now have the biggest and most unfair advantage that we've seen in the generative AI era. Claude doesn't come close. Gemini doesn't come close. Even ChatGPT itself doesn't come close.

OpenAI's Codex Desktop Program is the AI that the world has always wanted, but well, apparently they're too afraid to touch it because of the word code in the name. So we're going to break it all down for you today on our AI at work on Wednesdays series as we break down Codecs for Beginners Part 2 in our two-part series. And this is one, trust me, you cannot miss.

So here is the big picture. Codex has been the best kept AI secret for the past few months, but people are finally starting to catch on. Right. Um, I'll say this. There was a ton of anthropic clawed momentum uh in December and January. And it was well deserved. And I think how things happen, I've noticed this that I'm in the bubble of the bubble when it comes to AI.

Even things that I see, I feel the rest of the AI world doesn't always see them until maybe a couple months later. Cause I'm one of the only people in the world that spends 10 to 12 days or 10 to 12 hours a day literally doing only this.

And it is coming, so you should probably get on it now. And here's what else is happening. All the big players are moving to desktop and Kodaks is by far the best. Uh claud code is good, right? The Claude Desktop program. But the problem is, well, it's siloed. They don't talk to each other. Even Google Gemini went desktop, you know, last month bringing Gemini to desktop. And yesterday with their anti-gravity two point oh.

Funny, like funny enough, I don't know if anyone has says this, it just kind of looks like codex. Maybe that's a good thing. I don't know, but codex power feature. Are the difference maker. So that's what we're going to cover on part two. We're going to go over scheduled automation, skills, plugins, computer use, and remote control. So

Stick with me for today's show. Here's what you're going to learn. Uh, we're going to do a very quick recap of the basics, what we went over in part one. You're going to know the reasons most of your normal chat GPT usage should probably just be done inside codec.

You're going to know how the ChatGPT mobile app is now the most powerful AI controller in the world. That one's not even close anymore either. And you're going to learn how to properly set up codecs to be your around the clock agent that actually does your work.

Sounds awesome, right? It's not too good to be true. I'm going to be showing it all to you today, live on Everyday AI. What's going on? If you're new here, my name's Jordan Wilson and yeah, I do this every day. Um, it's unedited, unscripted, just bringing it to you raw and real as it gets uh with our daily live stream podcast and free daily newsletter. So if you're trying to be the smartest person in AI at your in your company, trying to grow your career, well

Starts here with this live stream podcast. But make sure to go to our website at your everydayai.com. We're gonna be breaking down all of the details of today's show. It's gonna be a fast and furious one. Full of facts. I mean, I could turn this into like a six hour course. Instead, it's going to be like a 30 minute podcast. So make sure you check out the recap in our newsletter.

And we're gonna have all the other AI news that you need there. All right. Also, I went crazy overboard on this one, y'all. I'm not gonna lie. All right. So uh make sure to repost this. And uh not only are you going to get early access and first access.

to our co-work codex cohort. Still not sure what the name's gonna be. Could just call it CCC, right? Like our old PPP. Uh so we're hoping to launch that in quarter two. So go repost this on LinkedIn. I'll make sure you get uh early access to that. All right. And if you are listening on the podcast, the LinkedIn link is always in the show notes. So check out the show notes for today's show and you will find that LinkedIn link. But

I also went absolutely bonkers and I spent way too much time in probably a couple million tokens putting together the Codex Cookbook. All right. So this is some advanced tactics. We have, I think, 50 plus.

uh use cases by job type, by category, by skill level. I've been using Codex around the clock uh since it came out in February. And this is some of my the the best of the best stuff that I don't have time to put in today's show. So uh when you repost today's show on LinkedIn, not only are you going to get early access to the Codecs Cohort and the Cowork Codex Cohort, but also the Codex Co-Fo.

All right. Let's get into it. We're gonna start live. I'm letting you guys know this one could get messy because we're gonna be doing some well. We're gonna try to be pushing uh you know codecs to the limits here, but I'm gonna be sharing uh my screen and a lot of other things. So live stream audience, if you could let me know if you can see this here. All right. Uh and to our podcast audience, FYI. Um

The video version of this one might be helpful for you to watch. I know people are always like, put your video on Spotify. Um I I don't know why I don't. Maybe I should have Kodak do that. But to go get the video version, just look on our website at your everydayai.com. All right. So

We're gonna do this live and here's why I do it. I don't edit this show. If it breaks, it breaks. If it blows up, it blows up. If it works, it works. Great. But there's a good chance it might not. The thing with generative AI, you get different things every single time. Um and I like to show y'all the, you know, the the real, right? Not just the polished version. All right. So here's what we're gonna do. I'm gonna go ahead, I have this queued up.

in my Codex now. So Codex is a desktop app. It is literally free to use, right? Your limits aren't going to be good, but I've already said this on the last show. Even if you have a paid Chat GPT account,$20 a month. You're gonna get a ton of codex usage. Uh, I've done testing. You get more out of a twenty dollar a month codex plan or chat GPT plan than you do a hundred dollar a month um anthropic plan. So um

It's a good deal, but yeah, if you're like, oh, I don't have a codex plan, it's not a separate plan, go use your paid chat GPT plan. So you download the app. I am on the Mac, but they also have this for Windows. It is a desktop program. You connect one time with your ChatGPT account and you're good to go. All right. So I have a prompt uh dialed up here. Let's see how it works. This is um This is hefty. Okay. This is hefty. So let's see. I do have a lot running on my computer, uh, but hopefully.

I'm gonna read this off. I'm not gonna read all of it, but I'm gonna read most of it. Here's here's I have a problem. Okay? I have a problem. You've probably noticed lately, I haven't done a lot of guest shows. Um, and one of the reasons was uh we had too many guests pitching. Um, some of our guest shows unfortunately ended up not being the best.

People just sold way too much, even though I had a hard no sell policy, right? And I got so many emails and they're like, This guest stinks. Don't do guest shows. So

put off doing or they're so time consuming or they're hard to do because I'm telling you my pain point right now. I'm telling you my story and how I'm, well, automating this in codecs, right? some new features over the last week or two have now gotten to one of my pain points that I don't like doing. I can automate a big chunk of it. I still keep myself in the loop intentionally. I could automate 100% of this, but I don't want to. All right. Anyway.

I get too many guest pitches. They come all over. They go to my personal Gmail. H have no clue how it gets there. Uh, you know, my old company called Accelerant Agency, I have a couple info emails. People guess my private email that I just set up and I'm Transferring everything over there. Anyways, I I get guest pitches from about six different places. Um

So different emails. Uh some of them are automated into spreadsheets. I have a form on my website. So it's it's it's a mess, right? It wasn't like this three years ago. I was begging for anyone to be on the show, but now we get thousands of guest pitches a year. So it's hard. Um also it's a lot of time to research people. Obviously, in people's guest pitch, they're like, this is the best thing since sliced bread.

Um, but I what I generally do is well, I look at people's background, you know, I'll use obviously AI to do that, but it's still a lot of these these steps right where I'm having to copy and paste a bunch of stuff and then I get lost. I get distracted, right? And I go down a rabbit hole and whoops. I just wasted an hour seeing if this person should be on the show or not.

All right. So here's what I have codex doing. So uh I already have full disclosure. Uh this isn't for vetting guests, it's for uh researching. So kind of our new guest format, it's essentially like

Uh let me know which one of those sound good anyways. So here we go in codex. I said please look at my Google Calendar because I do have three guests booked for next week. These are pre-recorded. I want to make sure they're good before you know we start doing these things live, get the format worked out, all that good stuff.

So I said, please look at my Google Calendar. And these things that are highlighted here in the Codex prompt for those of you uh watching along, these are technically plugins slash apps uh inside codec. All right, so I said please look in my Google Calendar and find my everyday AI guests for the next week. Um, then carefully search for them in my Gmail.

uh including inbox outboxes there's a lot of back and forth right i'm skipping over some of this um i need to find their original guest pitch info which may be in the guest pitch form uh google sheet in my google drive So so far I'm having it look in my Google Calendar, my Gmail and my Google Drive and kind of triaging or triangulating all this information between them. All right. And then I'm saying, uh most importantly, oh, I made a mistake, y'all. Um

Well, this is good. You're gonna learn something uh as we go. So I'm gonna um

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 six ninety one. Number two, tap the link in your show notes at any time for the start here series. Or you can just go to starthearseries.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.

All right. Uh and then I'm telling um Codex to go find the most relevant, popular, and aligned shows that might overlap with the guests based on their experience that they pitched. and research what made those shows successful and why. All right. And then I'm giving it essentially uh six different tasks to do. Okay, so first I'm just telling it to do extensive research on the guest background.

Uh, and I'm saying come up to with five to seven specific follow-up questions to dig deeper based on what they pitched and based on what's working recently in our analytics.

This is something again. This is my manual step that I would have to do every time. If I was the most focused, most locked in I've ever been, just getting through step the first part of step one, which sometimes, especially in like twenty twenty three when, you know, uh AI systems were were non reasoning and they weren't that great. Just getting through that first part of the first step would take multiple hours, right? This whole process might take a day or longer.

All right. So I'm saying come up with five to seven specific follow-up questions to dig deeper. Uh, and then I'm saying write a short and informal email draft in my tone style with proper formatting. So I'm literally having it go understand my style. And then uh go through, research all these guests, uh, research the best episodes that are most aligned with their experience on what they pitched, and then come up with very specific follow-up questions so the guests are more prepared.

Um and then I'm saying to start that draft in the email. That's not all. All right. Then I'm having it uh use the either the browser use or the Chrome to go into streamyard.com. That's what we use to stream. Uh I'm having it Click a new recording. So I'm giving it step-by-step on how to set this up.

All right. So it's like go click the new recording, uh, put the guest name as the title, click create recording, all this stuff. Um, and then essentially you're gonna get a link and then we're gonna send that to the guests themselves. All right. Uh then I'm saying

All right. Step two, uh coming up with three potential episode titles. Uh step three, I'm telling I'm saying to do additional research for me to prep for the interview. um such as highly relevant uh case studies, stats, facts, trends that are highly relevant to the suggested show. Number four, uh all right, and here's where I'm bringing in some other uh plugins or apps.

All right, I'm saying create two versions of a presentation prepping me. One in presentations, which will uh open as a Google uh slides file, and also one in Canva. I'm saying include everything I need to know about the guest, their background, the show type, the show topics, et cetera. Then I'm saying then create a simple web app using the build with apps skill. So this is a uh

Actually, I'm trying to remember if this is a skill or a plugin. I'll I'll have to check. All right, in case that is more convenient for me. Then last but not least, I'm saying also save the text only recap as a word doc in Google Docs. as well as a suggested outline for the interview based on what might might what might work best. So this is all. You know, especially when I would have a big guest, right? Like the CEO of a Fortune 500 company, uh, you know, back in 2023, just to do this.

these steps, right? I wouldn't create a web app and all these things. I'm just trying to show off showcase some of the capabilities. Um, but you know, it would take me a half day, sometimes longer, uh, to do this. Now I can, well, we'll see how good it is, right? We're doing it all live. So you'll see right now uh on my screen, it's already going through, it's already working. Um, you know, I had to interrupt it midway through uh with the stats that I forgot to attach and it's still going.

And then on the right hand side, uh it's showing me the progress. So uh at any point you can see the progress of codex because it's An agent, right? So by default, it breaks things down into multiple steps and you can kind of track it as it goes along.

So I can both follow along and see what it's working on, you know, what websites it's going to, uh, what files it's it's pulling, you know, out of my Google Drive. I can see all of these ha things happening in real time. This is literally like giving this project as if I just spit it out.

to a team of 10 people on the computer, uh right. And they're in a circle. And I just sit there in the middle of the circle and I look around and see what everyone's doing. I peek over their shoulder. This is the best way to get better at AI and specifically codex because it does work a little bit differently. Then you know, even if you're using right, I'm using uh GPT five point five extra high. So the best version.

Um, and I think I have it on, yeah, speed. Okay. I did put it on 1.5 speed. So using more credits, but the credits on codex, I mean, it's like free candy. It's it's free candy at a parade. I I literally can't hit my limits and I'm running it around the clock. Anyway. Uh I'm on the the best model and running it to 1.5 speed. So hopefully we can see the end result here. Um

But it still works a little bit differently than, you know, working with GPT-5.5 in ChatGPT would work. So it's important to have this kind of visibility and observability as you work. So you can see the progress on the right hand side. It broke broke this big task. down into uh looks like seven, uh seven or eight smaller subtests. Uh and then outputs. So this is where the final uh kind of artifacts are because I believe I asked for at least six things.

Uh and then right now it's the different sources. So for the sources, uh, these are all my connected plugins. So it's actually pulling in my dynamic data. So uh my Google Drive, my Gmail, Canva, Google Calendar, and web search. All right. So we're gonna let that cook and then let's get back into a quick recap, right? All that. Uh, but quick recap of part one. All right. So this is part two where we're gonna get uh go over some of the more advanced.

um techniques and not even super advanced, right? I'm not even gonna get into you know uh customizations and you know doing different things in the diff viewer and uh MCP right I'm not even getting into that. This is more the more advanced uh tactics for beginners. Okay. So this is still for beginners. But in part one, uh we cover this um the basics. So that was episode 776. Uh so last Wednesday. So make sure to go check that out. But here's the very quick recap.

alluded to the fact that codex is that. So uh if you want to be on the future, codex is the future. So right now codex is positioned as the operating layer for repeatable workflows, not just a chatbot interface. That's the biggest thing and we're going to see that in today's episode. All right. And a lot of people are instantly going to go to Claude. Right. And let me just say, Claude is

like the minor league baseball player right now in Codex is the MLB all star. They're not in the same league. They're not in the same division. They aren't. Right. I'm not biased when I say that. I I Go back and look, right? When I was covering, you know, Claude Code, Claude Cowork in, you know, December, January, February, I said at that time, cowork, you know, is in a league of its own. And it was, but it was clunky. Um, and it at least what we have today, not comparable.

So some of the biggest differences, and I did go over this a lot more, but Claude Desktop is siloed. All right. It's essentially three different programs that don't know the others exist. So you have to choose between chat.

uh or cowork or clawed code. And as an example, chat doesn't know what you're working on in cowork. Cowork doesn't know what you're working on in clawed code, and vice versa. And to make matters even worse, and I demo this live. Even if you go in clo uh cowork or code as an example and you do, you know, new thread, new thread, new thread, right? If you're on thread five.

It doesn't know anything. It only knows thread five. It doesn't know what you talked about in thread two. It doesn't know anything else. So not only is Claude Desktop siloed, but it doesn't have working memory in cowork or code, thread to thread or task to task, which is absolutely bonker. bonkers, right? When I would like when I do trainings and I tell like enterprise clients that and they're spending a lot of money and they all scratch their head and they're like, wait, what?

Yeah, that's the reality. And so uh codex is unified. It is one. There's not three separate entities and it has memory. So I did demo that uh for all you naysayers, uh, I did demo that live in part one. Uh and then last but not least, Codecs and Chat GPT. Well there's different use cases. I still do use Chat GPT on the web for some things. I'm using the uh mobile app now.

Around the clock, more on that later. Uh, but there are still some reasons to use ChatGPT on the web. For me, I'm using it for GPT-5.5 Pro because I do feel GPT-5.5 Pro is a little bit better than the 5.5 extra high in codec. Um, I also use it for workspace agents, which are a little better.

Not bad or different. All right. So uh for these codex automations uh that we're gonna run, codex has to be up and running, but there's all these settings that keep it on. Uh and then I use it for deep research. So At least for me, uh there's three things I still use ChatGPT on the web for because I know that's another thing people are confused about, like, oh well, should I just move everything to codex? For me, I've moved 95% of my ChatGPT web usage into Codecs, and I haven't looked back.

All right, so the five advanced features that you need to know that we're going to be going over, uh schedule automations, skills, plugins, uh, computer use, and remote control. All right, let's go ahead and reshare our screen. So uh we'll check in.

After I looked at it, I'm like, oh, we should change a couple of things here to make it better. So technically haven't run this, uh, but we'll see how it goes. All right. So I'm gonna go ahead and open a uh a new window of codex. Yeah, you can have as many windows as you want. I love using computer use and having codex control codecs and codex control claud code and all these things. All right. Uh anyways, let's do a quick overview of these five features. So first Let's look at plugins.

Uh so uh on the left-hand side of your codex app, you're gonna see uh some different pins, some paints. I went over the basics of this, but your settings at the bottom. Uh you have all of your projects.

um that are tied to a folder and they can read and write, right? That's another huge advantage of codecs. It can read and write uh files, which is can be good or bad if you don't know what you're doing. And then you have chats and then these chats are kind of individual and they're not tied to a certain project. All right. So and then up in the top, you have your plugins and your automations. So let's look at plugins.

There's no dedicated space for skills. So sometimes people are like, where are skills? Well, you have to click on plugins first, then click in skills, but we're going to talk about those in a minute.

Uh so plugins. There are I counted these earlier, but I forgot. Uh there's uh probably about sixty or seventy, uh, maybe more, maybe uh actually closer to a hundred. Um skills uh or sorry plugins in here uh so you can search for them or you know there's featured uh there's coding there's design there's lifestyle there's productivity

Um, and there's research. So uh very easy to install. Um, you would just click on something. So, you know, linear. I don't have linear setup. I can click on linear. I can just go click add to codex, and then I would log in with my linear uh credentials. Uh usually it's one or two clicks and then you're set up. All right. But let's let's uh a little bit more about plugins. So essentially

Chat GPT changed the name, right? So originally there were connectors uh in ChatGPT and then they turned into app. So now they're plugins um inside uh codecs. And the reason why is because they're a little bit more than connectors and apps. And let me show you what I'm talking about by going into the Gmail connector. Um so

Uh let me zoom in here. So the description is says read and manage Gmail. Use Gmail to summarize inbox activity, draft replies, and organize email threads through the connected Gmail app. So includes, and here's why. Plugins are not exactly the same thing as apps in ChatGPT or as they were previously called, connectors. That's because they also include skills. Not all of them do, but many of them have pre-package skills. And we're going to be talking about skills now.

All right. So uh these plugins are essentially connectors that combined skills and they link codecs to the tools that you already use. Uh like I said, there's a ton of popular ones. So, you know, everything from computer use and controlling Chrome, which we're going to talk about later. uh to spreadsheets, Slack, Notion, Linear, Gmail.

uh you you know your calendar both in uh google and in outlook versel sharepoint figma i mean hugging face netlifly i'm just you know scrolling through naming some of the more popular ones twilio remotion canva hey gen

Uh let's see what else do we got here? Obviously docs, atlas, and stripe, box, jam, uh, I mean bangers here, click up, uh, conductor, so many I use, fixer, fireflies, granola. HubSpot, Monday.com, Pipe Drive. So you have your uh CRMs in there, uh Semra.

Streak, Zoom, right? Something a lot of people use, like half the world. Uh Asana, right? So yeah, that's that's some of the more popular ones. So these are a lot of the day-to-day tasks that you use on an ongoing or the day-to-day apps that you use on an ongoing basis.

And if you didn't see the visual, right, you can use multiple of them at the same time. When I kind of read out that prompt, I was using, I think about six different apps. Okay, so um these are all, you know, permission is case by case. So you should go see and understand what they can do. So as an example, if I go back and click on Gmail, uh, so it includes the Gmail app and then two different skills. Um, but I can go down, uh let's see.

I always forget where this is. Where do I find this? The Gmail app. Website privacy policy. I had all this up beforehand and I'm like, oh, I should okay, there we go. Um, so you can right-click uh or you can uh click on the See, how d how did I do that? They should make this a little more clear. Okay, you just click on it, but you don't know you can click on it. Anyway.

uh you can click on it to see all of the different permissions or all of the different things that it can do. So in the Gmail's uh sense, this is different than in ChatGPT because uh, you know, early on in the connectors and apps days, they were one one way streets. Now they're bi-directional and they can read. Uh so in this case, the Gmail app has 24 actions. Uh 11 are write and 13 are read.

Uh so it can um, you know, go through, create drafts, create labels, all these things, uh, but it can also read emails, right? So it can write emails, read emails, do all these batch tasks. So You can click on any app and see exactly uh what it can do by looking at those different permissions or different tasks. And this is important too, because you know, if you're just giving very vague prompts.

to uh especially to these plugins that have a lot of different read-write actions and you're not watching the chain of thought, you might not get the desired result. That's why number one, it's important to look at the chain of thought because you will see uh essentially the the the tool calling that's happening there, um right, you'll see if it's oh it's it's grabbing the right uh permission or the right um you know tool out of those 24 from Gmail as an example.

All right. So that is plugins. So um I mean the big upside here is you know, no more bouncing between

All right. Next, let's go on to skills. So again, to get to skills, a little confusing in the left hand pane, you're going to click plugins and then there's a little spot up top. A lot of people like struggle to find it and they're like, where are the skills? Well, there they are. They're at the top.

Uh so skills are essentially like reusable instruction. uh and they act kind of as recipes. So it's like teaching your agent exactly how you want a task done every time. So uh that's in simple terms. Um in technical terms it's a series of markdown files, right? That's all it is. So you can type a forward slash command and then the skill name and then your full set of rules formatting and logic loads instantly. So it's

Why they matter is well, you get the same high quality output every time. So you're not going to get the exact verbatim, but you're going to get the same quality every time. So this is like uh think of it as like a series of prompts, um, very specific. uh well written prompts over and over, right? But that's all skills really are. Is there very specific questions?

format it in a um meticulous way that ensures between the different, you know, agentic uh nature of these models, between all the tool calling, uh the harness that's always changing, it makes sure that it really does uh what you want it to do consistently each time. And the easiest one to build one inside uh

Codecs, well, there's a couple ways. Uh you can go into the skills. Uh you can manage the skills that you already have. Looks like I have about 56 of them here that I use. It's actually not a lot. This is one of one uh this is one of my uh dummy accounts that I don't use a lot. This is just for kind of demos. Uh but you can in the upper right hand corner, you can go to create a plugin or create a skill. So then you can create a skill. There's actually a skill called skill creator that you just

to. So you don't even have to be writing anything in code. It's just in natural language. All right. So those are skills and why they're important.

All right, next, let's go to computer use. This one's fun. All right. Let's see what what could go wrong here. All right. Let's let's go ahead and try computer use. So I'm just gonna say computer. Uh so I'm gonna say use computer use. Um, and uh let's see. I'm gonna say open my um Apple News. We'll try that. We'll see. Open my Apple News and tell me what's trending. All right. Uh we'll do this. We'll make it Uh we'll go medium to go a little faster here. So computer use is actually that.

using my computer. All right. So Um, it's always funny when people are like, look, I'm doing this with no hands, right? But um what's what's gonna be happening here uh on my screen, I'm not gonna be doing. Uh the difference is. Uh the computer use plugin um is very, very good. Um I actually should have yeah, it's already done. Um I should have used uh Codex 5.3 Spark, which is amazing. It's so fast. That's the one with their partnership with Cerebras.

So um you'll see right here it's scrolling again, it's scrolling through um Apple News. So it launched uh a program on my desktop. It's going through right simple stuff. Um But it's so good. It's so, so good. So I say that because Claude's computer use went crazy viral when they announced it. It was a couple months uh before OpenAI announced theirs in codec. And I've done a lot of side by side now. Uh not only is

The codex computer use the default one faster and more accurate. When you use that 5 3 Spark Cerebras. I've done side by side. Um, in in in my experience, and it depends on what you ask it, right? Generally, I have it do very simple tasks when I'm doing these tests.

You know, open two or three programs, create a document, save it, right? Something like that, something that the average knowledge worker would be doing over and over. Uh computer use in Kodaks is about five times faster uh than computer use in Claude. Um cowork and if you use the cerebrus it's like fifteen to eighteen times. All right, so the Cerebrus five three Spark is not as powerful of a model, but if it's basic, if you're just doing basic computer use work, it is

absolutely mind boggling how good that is. All right. So that's computer use. So, um, right, it just so I click over here, it said, Oh, I Apple I opened Apple News trending stories and it recapped the six of the um uh top stories there. So um essentially it gives your agent its own cursor to look at the screen, clicks buttons, and types. And that's the biggest difference between uh the computer use via codecs and the computer use

via Claude Code. I think the Claude Code one has gotten a little bit better since they announced it. Uh, but for the first two-ish months, it was a train wreck. Uh, the main reason why is, well, number one, it was too slow. Uh, number two, if you were on the$20 a month plan, good luck even getting it to complete a task, you would quickly hit your limits.

Uh, but number three, it's screen jacked, it mouse jacked, it window jacked, everything. So you couldn't work at the same time as computer use. So You know, I was trying to integrate computer use into my day-to-day workflow, but it had to have its own computer. So I don't know what kind of wizards work over there in Codex. Um But it has its own cursor. And even if it's using an active window, you can still use your own window. So

If you've ever used computer use in Claude Desktop and you're like, oh, this is cool, yeah, go use Codex. Again, minor league baseball player. uh you know, Sammy Sosa. I don't know why I said Sammy Sosa. I was trying to think of, you know, an MLB All-Star from Chicago and Sammy Sosa came out. All right. So that is computer use. It is good. It has its own cursor. Uh, it runs in the background, no interruptions. Great. Uh, all right, let's go next. Scheduled automation. So

Now that I showed you plugins, right, which are essentially apps with skills, I showed you skills and I showed you computer use. What if you could just, I don't know, do all those things together and run them as a scheduled automation? Well, you can. All right. So um At any point, right, I could just do it here. I could say, you know, run this every day. So I'm just doing the the one that I was doing with Apple News. You know, I could say run this every day at um, you know, eight.

AM as a scheduled automation. Right. Um, that's all I have to do. So if you spend a lot of time in a thread and if you get something just right, you know, first say, hey, take a look at this, you you know, um, go step by step and Uh let's create a scheduled uh automation out of this to make sure that we get this kind of output every single time. Um, and that's it. So I just say create a scheduled automation and it's done. All right. And then you'll see up here, I have that new automation. Um

So this is called a heartbeat. So that means it runs daily. All right. And then I can go and click it and I can modify it if I need to. I can pause it. I can delete it. I can change the schedule. Or you can just go into the automations tab. So that's on the left-hand side. So you have your new chat, your search, your plugins, your automations, then you have your projects in your chat. Uh so I can just go into automation. I can just click a new one and I can also just build it from scratch.

You can have it work locally in a work tree on GitHub. You can select the different project that you want it tied to. So it can be it can just be a normal chat that's not tied to a project, or you can tie it to a specific project. Then you can set the cadence, whether you want it to run daily at a certain time, uh, things like that. You can also pick the model and pick the reasoning effort. And there are also templates, which are absolutely right. It's like how easy, right? Again.

If the reason you haven't used Codecs is because the word code, have you seen me write any code yet? Absolutely not. Yet I'm doing things that, you know, I'm accomplishing, I would say, uh a lot so far. All right.

Uh let's see here uh how we're doing on my other one. I have it open. All right. I don't know if this one is going to uh finish in time. Uh it must have been some of the um the new things that I added. Uh so Last but not least, let's go to remote control. So unfortunately, this is something I'm not going to be able to accurately demo because I'm using my phone to record this, right? Um

The this was just announced and now the ChatGPT app is, I think, literally the best app on the app store. If I'm being honest, I didn't really use it a lot. until the Codecs control came out. Um Mainly because I hate doing work. I've said this before. I hate doing work on my phone. I have fat thumbs.

Right. So if anyone out there, if you've ever texted me before, I'm not texting you on my phone. I'm waiting till I get home and uh office or whatever. If I'm on the go and you text me, I don't text. Right. I I hate it. But now I'm actually using my phone a lot more. And I'm using the ChatGPT app because now I can control um codec. via the ChatGPT app. I know this is a little confusing. It's not a Codex app. It's the normal ChatGPT app. Um, and it's a new remote control feature. So

Um, essentially you're gonna download or make sure you have the most updated version of Codex, the most updated version of the ChatGPT app, and there's gonna be a codex tab in the ChatGPT app. And then it's gonna take you through a pairing process, which is pretty simple. Um, if you get caught up, there's a QR code. Mine somehow just automatically just worked after I did that on each device and it worked. Uh so at any point, you can go here into settings.

All right. And then I'm trying to remember. I think it's under connections. There we go. Um, so as an example, I have I'm on my Mac Studio here, but I also have connected. Um my phone, so my iPhone, um, via the Chat GPT app. And I've also connected my um MacBook. So at any time, um

I can control other devices as well. So my MacBook is shut off and not powered on, right? But the cool thing is on my MacBook, because my Mac Studio is always plugged in, I can control my Mac Studio from my MacBook at any time.

Or vice versa, I can just control it via my phone. And that's the best part. All right. And let me quickly call out how again how minor league and the only reason I said this is because I think I like read some comments um on s the last Spotify episode and people are like, oh this guy must not must not use Claude. Yes, I use yes, y'all. I know every feature of everything. This is all I do all day, right? I know and understand that Claude, um has a similar feature to this called dispatch.

Uh but if you're talking about the equivalent dispatch um it's a similar-ish feature so you can control claude cowork okay claude co-work again it's three different silos from the claude mobile app uh button It's weird. It's just dispatch and it's one thread. So here's why that's absolutely useless. All right. It's essentially just a

I won't even say it's a two-way road. It's a piece of string. Uh, because the only thing it's really good for is if you have a file on your computer and you need it to send, right? Because you can't actually work on meaningful projects um in there. And that's the only

feature that the dispatch has to control Claude Cowork. It is not a full control your computer. It's you have access to one thread and you can go back and forth on that singular thread and that's it. But if you have everything else, all your other big projects in Cloud Cowork. It's not gonna do anything, right? It's just a dedicated single thread. That's not how it is in uh in codecs with the uh Chat GPT app. You can control everything.

any of your projects you have. One thing, there are some bugs, but the bugs on this are way less than the bugs in dispatch because I said this multiple times. My dispatch completely broke. I had to uninstall it. I think it's because I literally uploaded a um Uh I uploaded a uh Excel file that was like six megabytes and it crashed dispatch permanently. I had to uninstall everything and it still didn't work. Anyway.

This the the the bugs on the ChatGPT uh you know codex connect and uh connection app are very small. Like the s the biggest one I see is it sometimes won't sync an updated folder. Right. Big deal. Right. Cause I changed the name of a folder uh on the desktop, but when I open it up um on the iPhone, it still had the old folder name. But I still went in, I found the thread in that folder and it's up to date and I'm controlling it. So

Uh literally, I this is what I did last night. Y'all can I get super dorky? Um, I used Codex to open up. uh clawed in the browser. Cause again, you can control the browser, you can control your desktop, anything, right? Uh I had to go through all of my Claude All my chat GPD history and recap everything, put it all in a spreadsheet, um, and then write me like

dozens of skills and create dozens of new projects with custom instructions based on my usage, right? Talk about AI doing work for you. Uh that's it right there. All right, so let's see. Did we f did did this finish? Did this finish? Let's see. I don't think this one, uh let's see, it's gonna finish. It didn't. It didn't. It's it it it is gonna take it is gonna take a little uh a a little bit of time. Uh so let me just show you the the finished version here, if I can. All right.

So uh I did run this right before, uh pretty much the same prompt. I actually added two or three small things in the one that I just did and unfortunately didn't finish in time. So this one that did work successfully, uh, the other one will work. It's just gonna take a little bit longer. Uh that's the beauty of doing live demos of generative AI. So this one worked for 22 minutes, but I added a couple of things. That's why it took a little longer. All right, so if I go click out the side panel.

Let me do this. There we go. Um And I have all of my outputs here and all of the sources. So uh let me go to the bottom and let's just take a look at what this created for me. It has all of these key links. So let me see, click this. Okay. Obviously it's opening all this in a different window. Sorry, this is a little hard to jump around, but here's my uh executive readout. Very good. Everything I wanted.

all the different guests. My gosh, this is really good. Just one sheet that has absolutely everything. Uh love that. Here's the Google Slides version. All right. So it created a pretty, you know, basic looking, but It did the job. All right. Uh let's see the Canva doc. See if that works.

There we go. All right. This looks nice. Uh here it has the everyday AI guest prep for next week, the executive readout, uh, the Spotify analytics overview that tie to these different guests, the briefing on the three different guests, the current AI context, all this stuff. uh pretty good there. Uh the delivery index. All right. Here's all the different um links. So it did go through in StreamYard, created all these.

All right. And then it's uh I'm checking in another window here. If I can uh bring up these Gmail drafts without uh showing too much. All right. Yeah, I can't. I can't I can just take away this person's email. So then it won't matter. There we go. All right. Check it out there. There is the uh the email with the uh seven specific questions, the potential title directions, uh everything's in there, including the link. uh to the stream yard where we would record this. All right. So

Amazing. This is amazing. All right. And then a couple more deliverables as well. Here's a web app. It literally created, okay, look at this. It created an actual web app. Here for all of this, it's interactive. All right. So I can click on it. The overview, the different guests, the analytics that tie to these guests, the sources, the draft emails. My gosh, this is so good.

Uh the Word doc, I think I already showed this. Yep, there's the Word doc. Uh, and then also it created a PowerPoint. So it's previewing all these things FYI in the code. So Codex has a built-in uh web browser. So I can, you know, go and actually browse the web. And it also has a built-in file viewer as well. So I was actually looking at those. uh in the file viewer but at any time I can also open a browser uh as well. So did a great job. Like

My gosh. Like I can't wait to see when the other one does finish because I did throw a couple more things at it. Uh, how much better it is because I'm always iterating, always trying to make things better. All right. I know that was a lot, y'all, but We went live. Sometimes it works perfectly. Sometimes it's a little slower, but it's always worth it. But here's the thing.

It doesn't matter if a task like this takes 22 minutes or 42 minutes. And the reason being, is because these types of tasks, number one, I don't even have the capability. For me pre AI to code a little interactive website for all my upcoming guests, like it's cool. Do I need it? No. Like I can suffice with a Word doc or, you know, the little PowerPoint presentations. Those are helpful as well. But these are tasks that would have taken me

So that's why it's like, okay, if this thing takes 20 minutes, 40 minutes, it doesn't matter because I can have all of these things running for me around the clock. I can wake up to the work that used to take me weeks. Yes, I have to invest in the time. I have to stay up with the technology. I have to read the chain of thought. I have to iterate. But once I get it to that point, It's on autopilot, right? So now I'm probably gonna have this running uh at least.

twice a week, it's gonna look out two weeks in advance. And it's gonna do this exact same thing. I'm gonna go through, look at the next version, try to make it a little bit better, and then settle on it and say, okay, this one's good to go. And that's my new, my new routine is I wake up and I see that work that agents do for me. Y'all, I told you this when I did my 2025 uh you know prediction series. I said the future of working with AI.

Or the 2026. It is front-end orchestration and back-end tastemaking. That's what I do. That's what I want you to do as well. And that's where we're at now. This is what we've always wanted. Right? We've always wanted an AI that just does my work. This AI just did my work. Right. I have to provide the instructions on the front end, but it's cutting down the time that it would take for me to do this by at least 90 to 95%. And the output is.

better. Let's be honest. Because I'm looking at the chain of thought. It's going to sometimes dozens or hundreds of websites. I wouldn't have had the time to do that level of research in the high quality that it's outputting in all of these different artifacts that I have.

saved in there and now I can just have all of these things running for me around the clock. All right. So Was this helpful? I hope it was. This is one of those ones, again, you might want to go to our website, your everydayai.com, check the video version of this. And then if you are listening on the podcast, I always have in the show notes a little thing there that says join the conversation on LinkedIn. So go do that.

and repost this, you're going to get uh early access to our co-work codex cohort. Maybe we'll call it the CCC because that's a mouthful, as well as the Codex Coke Cookbook, y'all. Chef Jordan Wilson went absolutely. nuttier than a squirrel on keto on this one. The Codex Cookbook, it is going to be your guide. You are going to want it. I don't sell this stuff, y'all. I give it to you for free because I want to be your trusted source.

uh for knowing how to traverse the wild robe that is generative AI. So I could be selling all these things for a lot of money and people would pay for them because they're extremely valuable. This is hundreds of hours.

of my experience using Codex. I've been using the Codex app since the day it came out in February and I literally use it around the clock aside from when I restart my computer. So putting some of the best of the best stuff in that we didn't have time to to get to in these two parts, a ton of different use cases, uh, you know, filter it by category, by uh

level, right? A lot of different ways that you can use these kind of different uh recipes uh to get the most out of codecs. So I hope this was helpful. If it was, Aside from sharing all that, make sure you go to your everydayai.com. We're going to be recapping the highlights of today's show and a whole lot more. So thank you for tuning in. Hope to see you back tomorrow and every day for more everyday AI. Thanks, y'all.