Everyday AI の裏側：Jordan が最もよく使う AI ツールとワークフロー9選（再放送）

One of the most common questions I get is how I use AI every day. So on today's episode, I'm going to tell you everything. And the answer to that question obviously has changed drastically over the years as quickly as the AI offerings themselves have changed, which is all the time. But I feel over the last few months at least, my usage has gotten consistent enough that I feel I can make this episode and not have it be obsolete in like two weeks. because after using thousands of AI tools and modes over the years, you might be able to save yourself a few hundred hours and just copy some of my workflows and see what sticks. So on today's show, I'm going to give you a little behind the scenes on not only my nine most used AI tools and workflows, but also maybe some hot takes and trends on AI tool use in general along the way. So we're going to go inside Everyday AI and go over my nine most used AI tools and workflows that power my day.

All right, let's get into it. On today's show, here's what you're going to learn. My nine most used AI tools and workflows. You're going to know why I sometimes build my own AI tools, and I think you should too. and I'm going to tell you why I'm obviously codex-filled, yet still using Claude, Gemini, and other tools. All right, let's get into it. Welcome to Everyday AI. If you're new here, my name is Jordan, and we do this every day. It's for you. It's your unedited, unscripted daily live stream podcast and free daily newsletter helping business leaders like you and me not just keep up, but get ahead. I tell you what works, what doesn't. You take that information, all those saved hours, and you grow your company and your career. So if that's what you're trying to do, sweet, starts here. Make sure if you haven't already, please subscribe to the podcast and then go to our website at youreverydayai.com. The website is also always in our show notes there. We're going to be recapping the highlights from today's episode and giving you all the AI news and developments you need to know to stay ahead.

All right. We'll see if this one sticks, right? Watch. Watch later today. Obviously, it's Thursday. Big releases always come out on Thursday. They watch everything change today. But I feel, honestly, for the first time in a very long time, you know, my usage in terms of AI tools has not completely changed, you know, in three months, right? It's been fairly consistent. And some of those reasons is because I think there's, you know, one harness that will not be dethroned anytime soon. And kind of the move, you know, going from, you know, AI that was mainly web-based to AI that lives on your desktop and agents that live on your desktop, I think it made maybe the models less important. I think it maybe made some of the modes, you know, like your deep research and your, you know, plugins and artifacts, all these things, I think it made those things a little less important.

And honestly, between the models and the modes, I think there's actually a bit of parity, right? Whereas maybe two years ago when a new model came out, it was such a big gap. It was like first place and second place. Now, at least recently, between OpenAI and Anthropic, it's usually 1A and 1B, and then they flip-flop, right? But the harness itself, and this is why I think my usage has been so consistent probably since February, has made all the difference. So before I go into my nine most used tools and workflows, a little precursor first. Speaking of the harness, I'm obviously codex pilled. If you've been listening to the show at all since February, I've been saying this. I've called it. I said, Claude Code, it's going to lose its lead, right? Obviously, hats off to Anthropic. They really kicked off the agentic desktop era with Claude Code Desktop in December,

followed up by Claude Cowork in January. But they've just been getting smoked, and it's not even close in February. And, you know, I always will use the best overall tool no matter what, right? And, you know, throughout the years, I've had the most expensive plans to, you know, I was on the Perplexity $200 plan, you know, using their computer. I've been on the Google Ultra plan, I think when it was like $300 a month or whatever it used to be. You know, I currently have a Claude, you know, 200 Max and an OpenAI 200 Max plan. So I've used them all, right? And I still use all of them to this day. But the majority of my time now is really in Codex. It is in a league of its own. I don't think, you know, Cursor will catch them, although I am glad that Cursor added, you know, some simple Google connectors this week. I don't think Claude will catch. And also, I want you to first think about your workflow.

And the reason why I say those things is because I know our audience, you know, yes, there's obviously very technical people. There's builders who listen to the show. But for the most part, our audience is, you know, non-technical business leaders who are doing, you know, non-technical work. Right. So that's where my lens is coming through this. So, you know, if you're just like as an example, software developer, maybe cloud code is best. right? Or maybe cursor is best. But if you are someone that is, you know, not just 100% writing code, Codex, like I said, in a league of its own, and I don't see it being caught anytime soon. Okay, so with that all the way, I do want you to know that this is just based on my own workflows as well. Your usage is obviously going to vary and keep that in mind, right? So take that with a giant grain of salt. This isn't definitive, but I would say for the average non-technical AI user, like I'm saying, steal from my thousands of hours over the years of using all these tools.

When I'm saying, am I, you know, do I have a preference? Absolutely. Am I telling you the truth? Yes. All right. So just getting that out of the way. Also, another thing to keep in mind, my usage has really shifted. Right back in January, I was probably only 10 percent desktop, right? Because the only big desktop player at the time was, you know, cursor and CloudCode, right? So I was using CloudCode about 10% of the time and everything else was on the web. Obviously today that's complete opposite. So today I'm like 90% using everything on the desktop. So that's mainly codecs, but also using CloudCode, CloudCode work, you know, cursor, anti-gravity, Gemini on the desktop, right? For the most part, I am desktop agentic, harness through and through, right? Even for non-technical knowledge work, that's what I'm using. But like I said, I'm really only using, you know, Claude and Gemini Desktop anti-gravity cursor, so I understand them, right? So I have a comparison. And so when I tell you, yeah,

Codex slaps everyone else, that comes from facts, not just like a preference, right? So anytime there are new updates, I'm always using them, I'm always testing them, you know, doing some of my own, you know, internal benchmarks and use cases. So I know what's best for different types of work. The other thing to know is I am very bullish on agent control and agent co-working. And I'm going to tell you what those things are, right? I'm, you know, kind of taking you behind the scenes here at Everyday AI and how I work and, you know, how we kind of run things. But one thing to keep in mind is not everyone can work like this either. And I'm not saying that from a skills perspective, because it's not difficult. I'm just saying from a data and permissions perspective, right? I am having agents literally running around the clock doing everything on my computer, full control, full YOLO mode. I know that's not for everyone, right? If you work at a big enterprise, even if you've gotten permission to use quad desktop or codecs, you may not be able

to do a lot of these things. So keep that in mind. But if you are a small business owner, or if your company has taken off all restraints, and you can just do whatever you want, well, let's, let's have some fun. All right, here we go. I'm not gonna make you wait any longer. Here's my nine most used AI tools and workflows. And then we're going to dive into each one individually. So number one, I have my third shift stats agent. Number two, I have my third shift comms triage. Number three, I have my chat GPT sites to iMessage. That one's a little specific. Number four, I have my parrot dictation app that I built. Number five, I have my episode planning research. Number six, my AI app design. Number seven, Gemini notebook study, also previously known as Notebook LM, so did a show on that yesterday. Number eight, Codex Skill Building, and number nine, Remote Chat GPT Walk and Talk. All right, let's get into it. Number one,

third shift agents for stats. All right, first let me describe what third shift agents are. Well, they're agents that work third shift, right? Usually I'm up working fairly late, But let's just say my agents usually work about 11 to 5, right? Every single night, a lot of things are scheduled. And then I will also usually give it my toughest problem or set of problems to Codex every single night, kind of before I go to bed. I'll use sub-agents. I'll use goal mode. I'll use plan mode. All right. And if you're wondering, like, what the heck is all of that stuff? We did an episode. Let me bring up the number. let me bring up the number on that one. It was number... Come on. There it is. Episode 806, part of the Start Here series, where we went over desktop agent lingo simplified, talking about goals, loops, plans, sub-agents, and how those work in codecs and cloud code.

So go listen to that one. But essentially, my probably most used AI is not even me, right? I set these up. It happens third shift. My agents are working, you know, 11 to five. And then I wake up to things that are done. So let me tell you specifically what my third shift stats agents are. And again, I want you to think what just because I'm doing mine, right? Don't try to, you know, make my workflow fit your work. Instead, just see the path that I'm taking and apply it to yours Okay So essentially inside well even on the web but also on the desktop there certain you know pieces of software or apps that you use that connect really nicely to ChatGPT Claude, Gemini, Copilot, etc. And then there's some that don't. So for me, for everything that doesn't, right, it needs either computer use, browser use, or Chrome. Okay, so actually in Codex,

in ChatGPT Work, which are literally the same thing, right? So I'm using just Codex because it's just a little cleaner. And I usually use Codex, but, you know, it's interchangeable. So FYI, when I'm saying Codex, I also mean ChatGPT Work, which if you go to ChatGPT and download it, that's what you're going to get. So you can do the same things in ChatGPT Work, which is the desktop version of ChatGPT, right? But there's three different ways that these agents can use your computer in instances where there's not a direct connection or where there's not an MCP or model context protocol server. So in my case, right, we use Buzzsprout. So that is, you know, the podcast, you know, thanks to you all that listen. It's, you know, downloaded millions of times. And essentially, I didn't see this going in when I started the podcast. I'm like, no one's going to listen. And apparently, millions of people every year do. And so there's just some stats that I really need that just aren't readily available. by clicking export or by me going in and pulling up a couple screens on a couple monitors.

Like I have to do usually about 30 or so minutes of work every single day, just pulling these metrics in, exporting certain things, and it is time consuming and I don't like doing it. So this is my third shift stats agent. So it does a lot of comparisons. I have, you know, daily benchmarks, you know. So as an example, you know, our Wednesdays, AI at Wednesdays, you know, sorry, AI at work Wednesdays or working Wednesdays, you know, as we call them. Right. I know that every single day, you know, Thursday at, you know, 9 a.m. That shows, you know, should have like as an example, 11,500 downloads or something like that. Right. So I know every day that it can go in there. It can see if that is the case. It can go look at the last one from last Wednesday. Is it still getting a certain number? So I have all these kind of formulas that show me not only how popular a show is, but how does it do in its normal time slot or in its normal day slot versus others, right, for Tuesday and Thursday shows that maybe I don't have recurring shows on.

So I have all, you know, I had Codex make all these kind of formulas and it goes in, it pulls all these stats, you know, go look at day by day stats, which I can't export. Show me which, you know, shows are from two or three weeks ago are still maybe trending and, you know, getting two or three X what they should be getting at that point. So then I know that there's some evergreen appetite in that. And automatically I have a skill that if that happens, Codex is going to go start, you know, researching and finding more information on doing an updated show on that. So that's kind of my overnight agent. It also pulls information from our newsletter, which we use Beehive. So there is a model context protocol server for that. But even with that, there are some things that I can't get in there. As an example, we run an A-B test on our newsletter every single day. In the MCP, I can't get that. So I have to go in. It uses for that. It uses browser control. It goes in, clicks and finds all those things. So I have so many of these tasks that I used to do. Like two years ago, this is how I would start my day every day.

You know, and if I was very focused, I could get it done in, you know, maybe like an hour. Now it's, you know, 30 seconds to overview something, which is great. One quick thing, at least in Codex, it's a little different in Cloud Code, but in Codex, it's a little better. So they have a built-in browser. there is Chrome use and then there's also a Chrome extension and those two obviously work hand in hand and then computer use so you know the browser there's actually a file in kind of the Atlas browser was retired into Codex but there's a full-fledged Chromium-based browser inside of Codex so you know for many of these things I do them inside of the browser in Codex and I schedule them and they run overnight. Other times, you know, it might need to use a, as my second one here does, it might need to use Chrome because there's a Chrome extension involved. All right,

so that's my first one. But, you know, essentially it just helps me do multiple things at once. I don't want to go and click around. I don't want to, you know, copy and paste information in the spreadsheets. I don't want to be running formulas, right? I just wake up to an actionable briefing instead of spending my morning assembling one. All right, number two, again, this is another third shift agent. This is my comms triage. This one, my gosh, this is helpful. All right, and one of the reasons is I have, I don't know how or why, I have a personal Gmail. I have a Gmail account from my old company, Accelerant Agency. I have a Gmail and then I have two different Gmail accounts for everyday AI. And for whatever reason, people find all of them. Maybe they're connected to LinkedIn. I don't know. I get spammed all the time, right? We get, I don't know, more than a thousand people each and every year, you know, pitching

to be guests on the show, people reaching out about sponsorships, partnerships, most of those they don't pay attention to, you know, collaborations, whatever. So there's a lot of those. And then also LinkedIn, right? This podcast goes out on LinkedIn every single day. So people message me all the time. Sometimes I ask people to message me ideas, those kind of things. It's always me responding, right? But I do have agents go in and triage these things. So my second one is the codex triaging all of those things and telling me what's important, what I need to get to, and then also giving me draft responses. so I don't just have to sit at a blank stare email, right? And usually, at least within Codex, I have these things set up within projects. So in each of these projects, you know, there's a markdown file that I can change very easily, update very easily, that give a set of directions to these things. So it can work with all my context. You know, I can give kind of custom instructions to each of these different, you know, projects.

But this is good because, you know, in most AI systems, you know, Gemini, Claude, ChatGVT, Copilot, you can only really work with one email connected. So this one's really helpful for me. The way that I pull this off is, you know, just over time, I use the Codex record and replay skill a lot. But even this one, it doesn't always work. So I've had to really refine it over the years. But essentially, I have it launched Chrome profiles. and go to, you know, gmail.com for LinkedIn messages, right? I don't automate anything on LinkedIn because I don't want to get my LinkedIn banned. So I do have an extension in LinkedIn called Condo, essentially. So I have, it just syncs all my, you know, LinkedIn messages. It's a little quicker to go through. The LinkedIn inbox is so slow. So essentially, I have a skill go in. It opens up a certain Chrome profile. It goes to LinkedIn.com, clicks on messages.

That opens up a dedicated tab that has that Chrome extension that I need, which is why I do it through Chrome and not through the built-in browser. And then it can kind of go in, look at all my unreads, triage, find information there. So that's my second one. So essentially, I can wake up every single day. I know every single stat, right? We put out a ton of content every day. We put so much effort into planning the shows, into how we deliver them, into the newsletter. And, you know, if I didn't have any guidance, I would just be guessing in the dark, right? So it's great every day to have that in from my, you know, my third shift agents to have, you know, here's what performed well, you know, above average, below average. And then from a comms perspective, I don't have to go in and check, you know, four different emails and go into LinkedIn. It's all just there for me. All right. Next, uh, chat GPT sites to iMessage. All right. This one is great.

Um, if you're not using chat GPT sites, you are missing out. Um, Anthropic has a similar, um, a similar, uh, kind of option called artifacts. and there's live artifacts. I think there's great use cases for live artifacts. The reason why I still prefer Sights is it's a full, like, lovable, right? If you've used lovable, basically anything you can do inside, like, lovable or bolts or any of those kind of vibe coding full, you know, almost like replet as well, that's what Chachupity Sights is. I still think it is criminally underrated. It is so good. So there's so many things that you can do inside Chachupity Sights that you can't do in Claude Artifacts, right? One of those things is just better permissions. So you can make things public. You can force log in via ChatGPT. You can share things with your team. Not quite as robust inside Claw Artifacts. But what I love doing in this case is I have templates and skills that I've saved that make sites just really easy to use on my phone.

So there's certain criteria, both from a design perspective. I always have it built in light and dark mode. I always have it built in a font where you can make it larger and smaller. And then I always have it built in kind of like a bookmark thing. So, you know, all these agents, when they do report these things to me, yes, it will be in line. But oftentimes, I just have them update a pre-existing site. And then what I do is I save those sites to my phone. So it's almost like I have agents updating these apps. The great thing about ChatHPT sites is it has databases. It's like read-write. So I have one that's like a little basic CRM, and I can just open my phone every morning, see what's new. I didn't do anything. It's just like literally I have agents grabbing all this information, running it through my personal context, updating the quote-unquote apps for ChatHPT sites that I use. and then if it's a new one, it sends it to me through iMessage and it sends it to my phone, right? Which is great. Yeah, people don't know.

It's like that's a, you know, it's actually a decent way, right? I just tell it to go check text messages from Jordan and I give it, you know, explicit, you know, say hey it okay to text Jordan Jordan only It actually one of my AgentMD kind of instructions that go over all of my projects where it says hey you can open iMessage and text Jordan no one else All right. So that's the other one. I love this because I just get that continually updated personal apps without having to, you know, rebuild or maintain them. All right. Next one. I've never, I've referenced this like once, but I've never really talked about Parrot. Okay, so this is one that I built. And let me just tell you why I built Parrot, why I think it's better than other alternatives and how I use it. So Parrot is awesome. It's one of the first apps that I built. I use it literally every single day. If you've ever seen me, you know, do screen recordings, you know, on Wednesday shows,

a lot of times you'll see a little green dot at the bottom. Sometimes that's open. I usually try to like turn it off before I start in case I accidentally hit my dictation key. But a lot of times you'll see it on there. So if you've used something like Whisperflow or Super Whisper, there's all these platforms that, you know, they're good. But the cloud-based ones for me are just too slow, right? Even Whisperflow, I'd say, is the leader up there. It's too slow. It's not always accurate. And there's just features that just don't fit my workflow. And there's features that I want that aren't there. So I essentially built my own version that I think is much better. I built this with both Claude and with Codex. And it's essentially, it runs locally. So there's nothing in the cloud. It's stored locally on my computer, and that makes it much faster. I encourage everyone, if you want a fun little project to build and you're feeling somewhat ambitious, build a local dictation LLM.

So, yeah, you don't need Internet. You know, just tell Chachapiti exactly, you know, codex or cloud code exactly what you're trying to do. In this case, it, you know, finds the best text to speech or sorry, speech to text dictation model, you know, downloads it. But yeah, like for me, why I use this one, number one, it's faster. There's no lag. So I do like Whisperflow, but there's a lag. And then I thought, like, why should I have to send all this to the cloud if I don't need to? Also, Codex has great transcription and you can actually, people don't know this, you can use it anywhere, not just in Codex, but then the same thing. It's like Whisperflow, you know, it's there's still a little bit of a lag. So when I'm I obviously talk a lot and I dictate a lot, I don't want to wait a half second or a second for longer dictations. I need to go immediately. But then I have all these other things that I built into it. So as an example, I can type in any YouTube channel and it's going to every single time that there's a new video.

So, you know, I can have the open AI, Anthropic, right? Anything, any big company, you know, any other, you know, AI YouTubers out there. You know, I don't have time to watch anything. But if I want to know what people are talking about, it is literally every single video that comes out. It's going to automatically transcribe as long as I have the program open. And then I have a scheduled skill inside Codex that goes and it looks at all those transcriptions every single day. And it just gives me a little brief. Here's what everyone on YouTube is talking about. Right. I used to like three years ago, especially when I was getting started. I probably spent like 30 to 45 minutes every day on YouTube. You know, obviously watching it like 2.5x. I don't have time to do that anymore. You know, it's distracting. I just don't have the time. So I'd much rather just get, you know, a sites report. Here's what everyone's talking about. Here's the bullet points. Right. It's perfect. So that's a parent. It's probably my most used app that I've built myself.

All right. Next, I have my daily episode planning deep research. All right. So this is one of the few things, if I'm being honest, that I still use Gemini for. But let me tell you a little bit on how this works. So, again, this is running from Kodaks because it's the only one that can handle this. But a lot of times I save Twitter posts on my phone. So I bookmark them. Right. So if I'm walking from here to there, you know, whatever, at night, in the morning, you know, quick, safe things. So I save a lot of posts on Twitter. Another signal. Right. So it's all these signals that I'm looking at. Right. Like, what are people talking about? What's breaking? What's trending? Similarly, the same thing on Beehive. So that is our email newsletter. So again, I know like on our AI news section, if I put something in, you know, spot three, it should get this many clicks. If it gets above or below, I know that story either people don't care about or they care about it a lot. And then, so I essentially take those signals every single day.

Codex goes in, it opens a new browser instance. It goes to chattypt.com. It goes to claw.ai. It takes all that information, right, not just to save posts, right, but, you know, replies. I have it kind of do a sentiment analysis on anything that I save, and then I have it take all of that information. Codex is the kind of the agent orchestrator. Not me. Isn't that weird? I thought it would be humans, the ones that were orchestrating agents, but for me, it's actually Codex that's orchestrating agents more than I do. So it takes all of that information. it sends it to both ChatGPT, Deep Research, Claw Deep Research, and Gemini Deep Research, you know, to help me plan episodes. So essentially I say, you know, go find me, you know, five episodes, you know, go find all the information, you know, you choose whatever I bookmarked, whatever is, you know, well above the normal click-through rates inside Beehive, go do that in, you know, ChatGPT, Claw Gemini, take all that information, and then I have another

skill that helps me start public pointing and planning out different shows. All right. So now let's go to number six. This is kind of front end design for vibe coded apps. So I don't build as much as I'd like to, right? If I didn't have a daily podcast, I might build a little bit more, right? But I still, like I said, I have Parrot. I have another one I built called Planadoro. I have, what do I have over to my computer here? Codex Reset Watcher. I have a lot of like just Codex tools that I've built that help me use Codex a little better, right? But I have all these things, but I don't want them to be ugly. So this is something, if you spend a little bit of time on the front end, right, for even if it's a throwaway app, right, I build intentional throwaway apps. You know, like I said, if I have a little project, I'll build a little Kanban, you know, or I'll build a little app for a project that I'm working on. so in those cases it's like well it needs to look good because I'm not going to build something and have it be ugly so in that case kind of three tools this is just kind of like a three in one

here because I use them sometimes interchangeably sometimes together so that's Claude Design love Claude Design it is I think probably their most unique product so I love using Claude Design Google Stitch and GPT Image so using a combination of those just to create different front ends right So usually I'll have a couple pieces of inspiration, either things I've previously designed or just designs that I really think are good. And I'll put them into, you know, Claude Design, Google Stitch or GPT Image. I'll dictate what I'm trying to do, what I'm trying to build and get different outputs. So Claude Design doesn't do visuals, but it does great, right, mockups. It does great slides, et cetera. Google Stitch does great at mockups, right? It also can use Nano Banana, so it can do images. It can do website, everything. And then GBT Image obviously can do that as well. The thing I like really, probably a little bit more Google Stitch than Claude. So Claude, you can like export things to Claude code.

It doesn't really work because in most instances, it just sends HTML, which is not the best. I actually like using Google Stitch for that because you can then export the entire code of what it builds to like Google AI Studio. and then from there you can do pretty much anything with it. So this is just when I'm building front ends, whether it's for, you know, making changes, a design system for a new app, whatever it may be. These are the kind of my three ones that I use. All right, next, this is number seven, Gemini Notebook Study. You all know I just did a show yesterday. I use what was formerly known as Notebook LM, now known as Gemini Notebook, all the time. almost every single day. Sometimes, depending on the day, I might spend multiple hours, right? I like, even if I don't, this sounds bad, I don't always listen or use everything that Gemini Notebook makes me. It just depends on if I have time. But for the most part, I'll have it making me multiple assets every single day. So audio overviews, custom, you know,

slide decks, things like that based on, well, anything. A lot of times it's around episode planning. I do actually put a ton of work into episode planning. Ones like this, not as much, right? Aside from my nice, you know, GPD image codex design slides here. But I'm using Gemini Notebook all the time. So kind of like what I talked about earlier, how I do all these deep researches. I have Codex go and take these signals from, you know, Twitter, from our newsletter, run deep research on that kind of sentiment analysis. And then next I'll have a skill. So this is kind of like the combination or the continuation of that where Codex pulls all that and it creates custom notebooks for me. So, yeah, custom deep dives, audio deep dives, reports, I mean, infographics, especially if it's like I'm learning a new subject, right? Sometimes I'm learning at the same rate as everyone else, right?

Like when something new comes out, it's like, you know, learning about, you know, loops and sub-agents and all these other things, right? I'm obviously experiencing them firsthand, but I'm also trying to learn about them as fast as possible as, you know, from people who are, you know, who have more expertise than I do. So I love doing this inside Gemini Notebook. It's not just for me. It's not just a content generation machine. It is really interactive learning because I love my favorite one is probably the audio overview that you can do the deep dive and then do interactive mode. that's great because you can literally click like the join button and you can ask the two host questions so that like really valuable before you know a major interview or a project where I just need a better understanding a lot of times if I doing a quote unquote bigger show right I first have the deep dive, I guess, two hosts do it. And then I'll listen and, you know, kind of drill them like, hey, well, why'd you say

this that way? Isn't this a better way to phrase that or frame that? And they might say yes, and they might say no. So it's almost like sometimes I go through and I rehearse some shows. I'd say I probably only do that like twice a month, you know, when it's maybe something more challenging, something a little more new, a big interview, something like that. Then I'll go through and do it that way. But, yeah, that is my number eight. Number nine. Sorry, that was my number seven. So number eight is Codex skill building. My gosh, I'm big on skills. I think, you know, probably a year ago or, you know, even like December-ish, everyone was talking about lines of code, right, or tokens, you know, as being like a heavy AI user. I don't think that's a good metric, right? I think the number of times that you use a skill really shows if you're being an AI native person or not. So as an example, right, I have 124 skills that I've used and I've used those skills, a total of I got these stats.

This is just from within Codex from just one of my accounts. And I've used those skills 6,179 times. So I use skills, obviously, a lot, many times a day. Right. And I break up these skills into multiple pieces and then create a plug in. All right, so if you don't know what a skill is, it's just something that you repeatedly do over and over. And I want you to think of my codex use cases, right? Because I think a lot of people think of skills and they think of either just writing content a certain way or they think of coding something or they think of a design system that is in a skill, which is fine, right? The way I think of it, these skills, is taking a long multi-system, multi-app, multi-website task and breaking it down and turning each of those steps into a skill. And then in Codex, you can combine multiple of those skills and create a plug-in, right? That's what I do for a lot of things that I run and then I schedule those things.

All right, so let's say you go to, you know, Peck, I'll give you one of mine, right? So you can go to StreamYard. That's what we use for our live streaming. So you can go to StreamYard, download this, and then upload it into CastMagic, right? So we do CastMagic for transcriptions, right? That can be one skill, right? And then taking that and, you know, we take some of those pieces from CastMagic, you know, like bullet points of the show highlights, transcriptions, right? Then I can take that, copy and paste it into the back end of our website. That's another skill, right? So all these things that you do in succession, break them down into little pieces and save them as a skill. You can either do it after you do something, right? That's a huge, like essentially whenever I do something, quote unquote, by hand, that's not a skill. I ask myself, am I going to do this again? And chances are there's a decent chance that I might. So I create skills fairly, fairly frequently, right? I've used 124 of them.

I probably have a lot more that I don't use, but I've saved just in case. You know, I actually have a skill that runs every single day that creates new skills. So I have a skill that goes through. It looks at all of my threads on codex. Again, something that Claude can't do. Looks at every single thread on codex, and it's supposed to pull out and say, hey, these are things you're doing repeatedly, or this right here you should turn into a skill, right? You went back and forth six or seven times. A lot of times, especially with this new technology, computer use was greatly improved with the new GPT-5-6-Sol. But it sometimes makes your old skills faster. Or maybe a certain thing that works a little better before works a little differently now. So when I'm doing new projects or trying to automate something new, a lot of times I have to go through a lot of back and forth. And that's whether it's, you know, Claude, Cursor, Codex, Antigravity, Gemini doesn't matter.

That's just how it goes, right? But after you do that, that is your IP, right? That is your knowledge. And the great thing about large language models now is, well, just repeat it, right? That's the thing. People always think like, okay, well, I'm just going to do this every day and I'm just going to use these models. And okay, and I'm going to go in manually push the buttons. You got to stop being a button pusher. All right. Go and push the buttons once and then just have whether you're using GLaD, Gemini, Codex, just say, hey, turn this into a skill. Right. And it will. And it'll save the skill. But, yeah, you should be doing that as frequently as possible. All right. And then last but not least, my gosh, ChatGPT voice is great. insanely good it is so so good this is just another reason why i am confident like no one's

going to be touching codex anytime soon um it is so far so far ahead and if you haven't used chat gbd boys y'all go go and look online all the people that are you know normally even claude fans and you know bash open ai chat gbd whatever they will like go go find those people you know those people. Everyone's just starting to talk about this like this week, right? It is that good. It is. I did it show. I said, this is the next frontier of agentic work. So if you don't know what it is, right? I wish it was a little easier to get to in the app. So if you have the chat, you can be the app. Essentially, you connect it once to the, you know, either codex or, you know, ChatGPT work desktop app. From there on your ChatGPT mobile app, you open ChatGPT. On the side panel, you click remote. Then there's a little button at the bottom, a little blue button. That's ChatGPT live. So let me tell you what this is because you might say like, oh, doesn't Claude

have dispatch? Yeah, dispatch is terrible. It doesn't know any of your other threads. It only can work on one thread and it can't really do anything, right? Aside from you can kind of chat back and forth. ChatGPT voice in remote, literally, you can control your entire computer. Anything, right? So I can be on my phone and I just did this this morning, right? I've been taking more walks recently since ChatGPT voice came out. It's amazing. Thank you, OpenAI. I'm actually getting steps now, right? Some days I'm like looking at my watch and I'm like, oh, I got like 800 steps today, right? Now it's like I got 13,000 today, right? Because I get work done walking, right? And hey, in Chicago, we don't, we can't walk you around. So it's good now that I can go out and actually, you know, get some, get some vitamin D, um, and, and sort of sitting under the fluorescent lights all day, but essentially anything that you can do yourself on your computer, you can do it in chat GPT voice. So I can literally say, you know, Hey, chat GPT voice,

uh, look at my downloads folder. There's a new PowerPoint. Can you open that PowerPoint up? and then I have a new image inside Canva. Can you go download that and swap it out? Actually, tell me what's on page three. What's that image, right? It's going to tell you, it's going to literally, it's, what's the word? Why do I keep forgetting? Duplex, right? Full duplex, right? So what that means is you can talk to it, you can interrupt it, it can interrupt you, it can talk in here at the same time, but as it's talking to you, it has another model that works in the background. So, I mean, literally I'm just, just dictating, getting all these long, hard assignments, right? Hey, go, go into, go use this skill, right? Go, go download this from this website. I'm logged in, uh, you know, upload it here, change this, tell me what's on this page, right? I'm literally like, like I'm on vacation and I have an intern sitting in front of my computer. that's literally what I'm doing right and that is what you can do it is amazing and the great thing

is it's different which I know is confusing it's a little different than the GPT live mode in the chat GPT app this is the most confusing thing because if you use that kind of voice mode inside normal chat GPT it can't access your connectors it can't access your MCPs right that's the big difference. If you use ChatGPT voice mode via remote in the ChatGPT app, it can't. You can say, hey, what are my most important emails? Okay. Reply, say this, send, right? Like, hey, go, you know, whatever. Open my HubSpot. You know, what leads have moved? Okay, right. Okay, go update this. And then Slack, you know, my teammates, send them this. And it just does it all. You can literally do all your work in Chajabiti voice as I'm starting to do. All right. So that's a wrap. That's my nine. I just give you an inside look. Yeah, I gave you a couple of hot takes there, but that's just the truth, right? Might this all change in a couple of weeks?

It could, right? But honestly, I think, at least for me, I don't see my workflow changing anytime soon. And I think if you've been stuck on the fence, just take what I learned and run it through your own workflows, right? Run everything side by side through, you know, Claude Code and Codex and Cursor or, you know, through, you know, ChadGPT.com, Claude.ai and Gemini.Google.com, right? Do it for yourself. But I want you to take maybe my workflows, these nine that I shared with you as a starting point. Some are more habits, right? Like turning things into skills, but some of them I think you should just be trying, right? Like, you know, try using the combination of Claude Design, you know, Google Stitch and GPT Image. If you are trying to vibe code something, if you're trying to build a little app for yourself, your team, your company, right? Start there versus, you know, building it all and then saying like, hey, this is ugly. Let's change some things. You're going to save yourself so much time. So I hope this one is helpful.

If you like this type of show, let me know. Maybe I'll do another one or two. but that is it inside everyday AI and my nine most used AI tools and workflows if this is helpful please subscribe to the podcast then go to your everydayai.com sign up for the free daily newsletter thanks for tuning in see you back tomorrow and every day for more everyday AI thanks y'all