← AI-Driven Marketer: Master Practical AI Marketing Skills
AI-Driven Marketer: Master Practical AI Marketing Skills
Marketers, Stop Using ChatGPT. Start Using Codex.
00:44:21
Loading summary
Transcript178 lines
- [00:05]
A
It's been quiet on this show for a couple of weeks because everything's been changing and it's been changing for a few months. But I Trav, I think I finally had the epiphany of what was going on and I've been digging into it and I'm here to report on some freaking changes. I swear things have changed for me more in my how I'm approaching AI more than like the initial deep dive. I did like two and a half years ago when I first started the show. It's that dramatic, it's that different. It's not going back. I feel like I'm a little behind in it, honestly. I feel like some other people had figured it out. But now that I'm in, I'm diving deep and I'm going to reporting everything that I've learned on this show. So welcome back to the AI Driven Marketer Bot Bros series where we take the news and separate out the help from the hype. I'm Dan Sanchez and I'm joined by my brother, Travis Sanchez. I'm here and today there's a lot to cover because obviously we were dark for a few weeks. Sometimes you just gotta grind to get work done. But also, like, I think I just needed a little break to simmer down and just dive into the new way of approaching AI, which is essentially saying goodbye to ChatGPT, saying goodbye to Claude. I mean, I still have them around, but I'm just not using them nearly as much anymore.
- [01:16]
B
Wow.
- [01:17]
A
And I'm now just using Codex or Claude code, either one of those. Every time I say Codex, just insert Claude code in there and you could use either one. But I use Codex cause I'm a chatgpt guy though. I'm using Claude code for work because they, they, they gave me a premium whatever license and I use that. But I Even upgraded my ChatGPT from a Plus account to a Pro account for this because I instantly realized that I was going to be using it way more as soon as I had the epiphany of like, what it was and what it could do. I know it probably sounds to you like, what the heck are we talking about, Dan? How much better can it be? But I'm telling you, the only frustrating part about the whole thing is that it's not available on your phone. It's a desktop app. So I just use. That's what like I use Chat GPT mainly just because that's what I have available on my phone. I don't have Codex, but I wish I Had Codex on my phone, it would be better, but we only have chat GPT. So let me explain what Codex is and kind of let me back up a little bit before I actually unpack, like why this is cool. Codex is OpenAI's coding dedicated coding tool. Okay. You're like, great, we're marketers. This is a marketing show. What does this have to do with anything? Everything. It just so happens that we've been trying to figure out this whole agent thing over the last couple of years. Right? Right. As soon as we think they thought they had unlocked the ability to do agents, but when they had reasoning models and it could think through things and make decisions. And they had some reasonable success with that after optimizing. But what they found was that the AI that was like really fine tuned for coding ended up being the way forward for agents. And it's not because they could code, but it's the way you approach code that ends up being the unlock for agents.
- [02:54]
B
How you can code what you can code.
- [02:57]
A
Let me break it down this way. Have you ever seen, have you ever seen the code for software before? Maybe on access? Well, not even the code. Like have you ever opened up an app and then got into its backend somehow? Maybe for like some software program on your computer and you're like, oh, there's a bunch of folders.
- [03:14]
B
I think I have. And it just looks like a bunch
- [03:16]
A
of characters, folders and files with randomly gobbledygook in it. Right. Well, it's because the way you organize software, it's not like one word doc with all the code in one document. That would be too much and it would take too much to load the whole dock every time. So a long time ago, software developers got smart and they separated out into a lot of different files. This is from long ago. They've been doing this for a long. And they organize it in different folders. Like you have a folder just for different functions of the software. You have a folder just for the resources it might need for different things. You have a folder for utilities and all kinds of things. And they separate it because they have different little snippets.
- [03:52]
B
Speeds it up if it's separated that way. So it's not like using as much ram. It's a way to organize. So if you're like, it's not working in the, like whatever this particular function
- [04:03]
A
or feature isn't working.
- [04:04]
B
Got it.
- [04:05]
A
I can go and just troubleshoot that and it won't corrupt the whole thing.
- [04:07]
B
We're just scrolling through pages. Yeah, I Got it right.
- [04:10]
A
There's a lot of reasons why they separate out the files. Now when they started making AI specific for developing code and troubleshooting code and adjusting code, it had to get the ability to think about the context of what they were trying to solve across multiple documents. So it could go in and think like, oh, well, it's probably over here in this file. Let me check. Let me read this file. Okay, this file connects to these files and it would take in the context of all the things it would need. It would think and then it would go and look some more and then it would think and then it would go and look some more and then be like, hey, I think it's this, this line of code. Let's check. Is it that line of code? Test run it. Okay, that one's working pretty good. Now let's go make this change. So like it's thinking through a lot, but there's a specific way to working with software that ended up being really good for more than just software, because that ended up being the thinking process to navigate almost any project. Not just code projects, but any project, because it essentially gave it a mind to be proactive, to go find its own context it needed. For example, I built a process to do pre production for this very show. I took that process and it was at a custom GPT. I'm like, this is the steps. First I spitball. I just dictate an angle for the episode. I'm like, this is the angle I'm going to go for. Like the general idea. And it usually I have it. It's programmed, it's not program. I just give an instructions like, hey, first take my idea, give me five different angles to approach it for the show. And it does that. I pick one and then it gives me titles and it gives me captions and then we dial out the im. I took that into Codex and it took it and it said, okay. And then I gave it a bunch of different shows, a bunch of different files about the show. I just gave it a bunch of context. It organized it into a folder system. I was like, great. I'm like, let's begin. Let's create. Let's start with this episode. And it went through and started just finding everything it needed. And it actually found other things I didn't even think it would reference. It went and found folders and thought about how to incorporate it in different ways. I actually messed up the first time and then just deleted that chat, then restarted, but I forgot it already started creating files for the episode. So when I restart, it's like, huh, it looks like we already have an episode that we started on today. And in fact it looks really similar to the one we had already worked on. So I think I'm just going to take his new spitball here of dictation and it looks like he gave some context that was missing from the last one. So I'm going to add that to the document. You could see it thinking through the process as it was referencing and found the files from before. I'm like, chatgpt can't freaking do that. It just starts over again every time. But in Codex, if the files and folders are there, it goes and searches for the context every time, which means it builds on itself. No more. Do you have to like create custom instructions. And then if I wanted to change the process, I have to go and nitpick the instructions myself. I could be like, hey, actually even in this today. Because they even changed the. Changed a few things. I was like, hey, actually we really don't need to do captions and images separate now because of the new image generator. Let's do those together now. And I'm like, just update the process document. And it goes like, yep, updated the process document, done. So when I ran it again, guess what, it was already updated. I didn't even have to touch it. I just delegate and be like, oh, actually what I meant was this. Make sure that every time we run through this step again, just do it that way. Done. That's different. That's different, right?
- [07:20]
B
And the only way you can access this is if you're paying the a hundred dollars a month for.
- [07:24]
A
No, no, no. You just need the $20 a month. The reason why I upgraded to the $100 a month is because I realized I was going to be using this a lot more.
- [07:31]
B
So I hear this, I hear this, and I'm trying to like work through a process that I'm currently building out or needing to do. And I'm like, how do I experiment with this?
- [07:41]
A
Yeah.
- [07:42]
B
Where I don't get lost in the weeds. Or I'm like, I don't get it. Because you are a tinker kind of guy. Yep. And you have a lot of experience back end with building websites, running ads, building blog pages that have gone in the millions of views on a monthly basis kind of jam. So I'm like, how does someone like me who's an active AI user, but honestly on the gooey side, where it's like I wasn't using the computer, where it was like they're typing only and there's no screen. Like, I want. I want it to be frictionless and user friendly. What's an example of like, yo, go try and build this thing, or, well, here's.
- [08:21]
A
Here's a common routine task for any kind of communications. You probably do this even for the work you're doing now. Probably once a week, you have to do something where you have to produce an image or some kind of set communication. And it's the same thing every time. Right now you could make a custom GPT for this, but you could also do it better in Codex, for example. Like one example I'll give right now that was probably relevant to every single marketer. Every week, I pretty much have to come up with a promotional campaign for Trim, Healthy Mama, for whatever the fricking sale we came up with. Right? And it knows I need. I need a campaign. I need an angle, I need to know what I'm bundling. And then I need a set. I need, like three different images, very specific formats every single time. I need a Facebook post, I need an email, I need a text message. If I'm feeling like doing more, I'll make a couple extra social posts.
- [09:07]
B
Yeah, okay.
- [09:08]
A
Right. This morning I just gave it like, hey, I need these every time for one campaign. And I'm like, put that in the process, doc. This is what I need. This is a style. I need it in. After delegating to it, I'm like, okay, well, here's the general campaign. Run the process. Done. And then I went over and opened up a new tab. But because it has access to that process, and I ran through it again, I'm like, I need these again. Done to spec. And there was something wrong with that. The second time I was like, oh, in this case, do this. Don't give me the email, give me this instead. Done. It knows it's now watching for that use case. And so every time you use it, it gets better because the process becomes more and more evolved.
- [09:42]
B
Wow.
- [09:42]
A
So what communication are you doing every single week? That's like, like clockwork. What other clockwork things you have going on?
- [09:48]
B
Right now I'm in charge of recording media, social media videos, kind of like Billy on the street interviews at the church that I work at. So that's one communication. Obviously, we have a daycare attached and they need help. Like, they need marketing help.
- [10:05]
A
Yeah, hard.
- [10:06]
B
So I'm even just thinking about, man, what does consistent posting or even ad creation for them look like? And I'm like, it's a whole process.
- [10:14]
A
You write Facebook ads.
- [10:15]
B
I mean, they are, but they're terrible. Someone else is doing it, and it's awful. I see it when I scroll and I go, oh, my gosh. This communicates nothing that I would want for my child.
- [10:24]
A
So, yeah, so you can make. I'm using it to create Facebook ads faster than I've ever done. This is partly like the other Update. I use ChatGPT image a lot, but ChatGPT image is so much better in Codex than it is on the app, because in the app, it's slow, right? You make one at a time, and it won't let you start the other one, generally until the first one's done. On Codex, you don't have that problem. I could say, hey, come up with 10 different ideas for images. Give me 10 different angles for this ad campaign. After I've already brainstormed, like, the title and the angle and the copy and all that kind of stuff, I'm like, great, based on this. Give me 10 different Facebook, 10 different creative or wild or whatever you want it to be. You know, give me 10 ideas. It gives you the 10 ideas I always. I always love. Before I generate images, I have it ideate and text first, and then I. And Codex, you could be like, great, make them all. It takes a while for it to make, but it goes and just makes it. I go work on something else. I go get another agent started doing something else, and I come back and all 10 images are done, and they match the prompts, the vision, the text prompts that it gave me. Here's what's really wild. Now it can code still. So before, if you created all that content and ChatGPT was stuck in ChatGPT, right? And what if ChatGPT didn't have some kind of integration or connection to the tool where you actually publish stuff? Well, you. I mean, you. You're done. You'd have to go find some other AI tool that maybe integrated the thing you want. Not anymore. I literally sat over this weekend. I spent. I spent my whole weekend doing this on the side.
- [11:44]
B
Of course you did.
- [11:45]
A
I was like. I was like, chatgpt, how hard would it be to connect to High Level, which has a social media tool baked into it, like a posting and scheduling tool? I was like, can you figure out so that every time I say post it to social, you just go connect to my High Level, which already has access to all my social accounts, and you just post at all? It's like, yeah. And it looks up the documentation, writes the code. It took a while because they had to troubleshoot some things. Their API documentation wasn't very clear, but it actually dug into the forums and eventually figured it out.
- [12:11]
B
Geez.
- [12:12]
A
And now it has that code available. I don't have to touch the code. There's no user interface for this. But now every time I tell ChatGPT to go post it to my social, it's like, done. Because it already wrote its own tool and its own custom integration to do it. I'm like, yeah, post these today and then schedule the other ones for next week. Done. Scheduled in high level. I can go and review it and edit it myself. But I already have the posts in front of me and it's already done. Doesn't that feel a little bit more agentic? You can string them together too and be like, hey, every time you run this process, create these assets and then don't even check with me, just schedule them in this sequence.
- [12:46]
B
You know, you're talking about really strong use case scenarios of making something like that happen. If someone's listening and they're like, how do I even just do something simple where it's not maybe running code for APIs with a program they're using, you know? But I almost feel like you need to create a little course. Course?
- [13:06]
A
I don't want to do a course.
- [13:07]
B
It sounds like a strong word. I'm talking like, hey, first step, go open Codex and ask it to do this. And then see the difference between, you know, one instruction versus the problem.
- [13:18]
A
It sounds intimidating, but let me, let me just show you. Let me just show you.
- [13:22]
B
It's kind of intimidating, but it's hard. It's hard to like, visualize.
- [13:25]
A
I know, it's. It's crazy. Okay, where do you even reach Codex?
- [13:28]
B
You just reach Codex on the. Listen, I saw, I saw. When I've been using AI, they go, try Codex and it's like at the top.
- [13:34]
A
It's at the top or it's even in the left hand sidebar, like right underneath. It's like new chat, search chats. Codex. Codex will open up to its own web page. And then you will have. Have three options for Codex. Install the cli. Don't do that. Download the app. Do that one. And then there's the. Try the IDE extension you want. You do not want the left and the right one. Those are for like code power users. You just want the app. Get the freaking Mac app. It's smooth. It just looks like ChatGPT. But there's a few differences here. You're. I could tell you're you're already browsing and looking for it now?
- [14:07]
B
Yeah.
- [14:07]
A
Let me just show you.
- [14:08]
B
I just want. I'm like, well, how hard is this?
- [14:10]
A
You know, it's not. It's not hard. Just watch my screen that we're using Riverside. I can actually just share my screen because it's a lot of Riverdale.
- [14:17]
B
Sorry, this is a show. It's a dumb show on Netflix.
- [14:19]
A
So look, this is.
- [14:22]
B
Okay, you're in the. You're in the AI window.
- [14:24]
A
Yep. Can you see it? What should we work on? Right, and then it has the chat codec.
- [14:28]
B
Or that just AI or that's just chat.
- [14:30]
A
This is Codex. It looks a lot like chat.
- [14:33]
B
It looks a lot like that.
- [14:33]
A
There's one major difference. You got these little folders right here. So when you add a project, it's kind of like adding a ChatGPT project. Right. But instead of it being a Chat GPT project, you could say, add new project. I'm going to start from scratch. Okay.
- [14:47]
B
How did you get access to High Level?
- [14:49]
A
Through the API. It just coded its own access.
- [14:53]
B
You're talking, you're talking. I mean, I know what API is, but like.
- [14:56]
A
Okay, I'm going to say use an existing folder. Okay. These are all my folders. Okay. I'm just going to create a new folder on my documents in my computer called Test. Okay. Now I'm going to say, okay, now everything in this test folder, these are the. These are the files it's accessing. It's on my computer. It's working from the files locally on my computer.
- [15:14]
B
You downloaded this? Yeah, from the app store.
- [15:16]
A
Yeah.
- [15:17]
B
And it's called Codex.
- [15:18]
A
It's not. Sorry, it's not on the app store, but if you just Google Codex Mac download, it'll be that first. Okay, so now I'm in Codex, Right. And you could just. I created a new folder, but it's really. It's a folder on my computer locally on my computer in the documents. So every time it creates new image assets or whatever, it's storing it in that folder that you can access later. Let me show you one that I even just did for this very episode, right? It's just me chatting. Except every time I chat with it and it's updating files, it's telling me this is like a code commit, but it's really just updating the process doc. So let me see. This is the pre flight doc for this very episode. It open up here on the right, kind of like a canvas. This is a document on my computer now, but here's the workflow. I updated the workflow in this last conversation. I was like, all right, let's move forward with that. Let's add a final step to the pre production process and separate out the images from the group. Because I had it create a ton of images. Here's all the different types of thumbnails I looked at because it can create all of them now. So I'm like, great, make me just like nine different images. And then I just picked the one I liked. Okay. And I said, all right, since you can make whatever image now, I want you, once we pick a winning image, I want you to make a square version of it. I want you to give me a four or five vertical version of it. And then when we finalize the image, I want you to make a pre flight document with the episode image, title, frame description and opening hook that I can read that frames up the episode. So I'm like, I'm giving an instruction for the next time. But then I'm like, hey, after you finish updating the documentation, actually go ahead and run the play. Like run the documentation. So it updates it. And this is the whole, this is the pre flight. Where's the process workflow? There it is. Here's the pre production process workflow. It's just step by step instructions like you would give an employee to run. But it's for itself. It's taking notes for itself. The next time we do this, it's going to run this, this play. And it's smooth. See, it shows me all the documents that it created and updated. Like, here's the, here's the square version of the image, here's the thumbnail. Here's the social post version.
- [17:17]
B
That's awesome.
- [17:18]
A
And now every time I run this play, now it's gonna run it and if I want to make updates to it, it's gonna update it. You can see it's like this is a very code level thing to do. It Change 29 lines, remove 15 lines. Because it's updating the documents. So if you can see by this conversation I'm just kind of scrolling through for those listening, it looks just like a normal chat application except way more robust. Yeah, it's a little bit more robust because it's, it's trying to talk to you and tell you like, oh, here's the link to the new workflow doc. Here's the changes I made. It kind of shows you, like, here's how many changes I made to the document.
- [17:49]
B
So if somebody is working through what are all those social Media posting platforms like, oh my gosh, I can't believe I'm. My brain is blinking route or what are all the different ways people post
- [18:01]
A
and schedule posts like Hootsuite and Buffer and all those things? Yes, yes.
- [18:05]
B
If they wanted to connect this to one of those platforms, would they be able to?
- [18:10]
A
Yep. So let's do it. Like this is the conversation for the project build out.
- [18:14]
B
You can go to Buffer, take their API, plug it into your folder here of like social media posting. Bro, this takes. Dude, you're on, bro. You're on a different level.
- [18:24]
A
Yeah, okay. I'm trying to show you the future. Everyone's going to be doing this right now. Right now.
- [18:30]
B
No one's doing it right now.
- [18:32]
A
Some people are doing this right now.
- [18:33]
B
Yeah, but how many people? Like 01% of AI users more than you would think. No, no.
- [18:41]
A
Yeah. In fact, it's, it's part of today's poll. Do you want to see? Because I asked people how many people were doing it. So hold on, let me share a different version of my screen. We're going to skip ahead to today's poll because I freaking asked people and you'll be shocked by the results. Of course, this is a more technical crowd. This is how I knew I was a little bit. Because even my own audience was ahead of me on this because I hadn't been doing this yet. I think I'm quickly accelerating past everybody else. So like everybody else is dabbling and I'm like going into deep mode really fast. But look at this LinkedIn results. Are you using Claude code or Codex or. Sorry, Claude code or Cowork or Codex to regularly for non code tasks. And 31% said all day, every day, 28% said daily, 9% said weekly, and 32% said rarely, never or not. Sure. That's a lot of people using cloud code or Cowork, which is kind of like Claude code light. I don't, I don't understand Cowork. They should have just. People just need to start using cloud code or codex. I'm a ChatGPT guy and I just, I use the images so much that it just makes sense for me to use codecs and I can't find a problem that Codex can't solve for me. So I don't really understand why I need the extra power of cloud code because I'm not really coding really robust applications. I'm just doing lightweight code work if at all. So I don't like. I don't. Codex works pretty dang good. And you Get a heck of a lot more.
- [20:03]
B
Wow.
- [20:04]
A
Well, so like I said this, I'm telling you, the way it works and the way it operates is game changingly different. I know it sounds weird now, but a year from now it will be standard. Just like a year ago, you know what we were doing that we weren't, that we just started doing? This time last year we were using O3, which had just dropped. And the big difference is that it could do searching by itself. It could go search the Internet and it changed the game. Remember we had reasoning models and we
- [20:33]
B
had search but it couldn't, it couldn't
- [20:35]
A
actively prove what it was finding. And when they dropped O3 and the reasoning model could go and look on the Internet and verify and think and then give you back customized results based on what it found. It was like mind blowing. And now it's like standard. We don't even think about it. How mind blowing it was a year ago, right? So I'm thinking this will be standard practice a year from now and people won't even think about it. This will just be. The new way of operating is using code tools. They will probably rebrand it away from being a code tool. We might even still call it Codex. But we won't even be thinking about code. It'll be writing code and we'll all be coding. But like it just won't even be a thing. We'll just be delegating. We'll just think about it like delegating.
- [21:16]
B
Wow. Well, you're. This is like next level Left brain deep dive 4 hour focus. Built something on the back end where
- [21:26]
A
I'm like, I know bro.
- [21:27]
B
I'm like, I'm the front facing guy
- [21:29]
A
on the camera, bro, but trust me, install Codex and just start using it.
- [21:35]
B
Like, yeah, yeah, I downloaded it. I downloaded. I'll try and download it.
- [21:38]
A
And then every time you want it to remember something like, hey, every time I ask for this, remember this. Like document it. It'll do it. And every once in a while just give it one folder to work in at first and just be like, hey, re. Reorganize the folder in a way that makes more sense. It'll just reorganize the whole thing. Move files around, change folder names and all that kind of stuff. Shoot. You can just give it access to your desktop and be like, reorganize my desktop? It'll start doing it and renaming things. It'll be like, oh, but run it. Run it by me before you make the changes. It'll be like, hey, I recommend Doing this, you're like, do it. You can give it pretty strong access. There's different levels that you can give it to. Like, you can say, hey, every time you make a substantial change, ask me first. I just turned that off. I just want. It was asking me for permission for so many things. I was like, screw it. I'm just saying yes every time anyway. Just. But I do limit its access to particular folders so it can't go outside the folder and accidentally swipe something that maybe was secure or private. I don't know. So that is something I'm doing. And I find, like, for what I needed to do, I don't really need it to be a go beyond any particular folder I give it access to. If I need it to have access to something, I just drag and drop it in the chat. Be like, oh, take this into a fur account. This is my, my style guide. Use this every time you design an image, reference the style guide. In fact, come up with some documentation for what's in the style guide and how to think about it. Take the. Pull the colors, pull the fonts, all that kind of stuff. And now it is. So that's some big news. If you're listening to this and you're like, what the heck? Go. Trust me, Just go install Codex. If you're a ChatGPT person, go install Codex. If you're a Claude, download their app, switch over to Claude code, and just start using it. It's. It's a little. You just have to start forcing yourself to use it and then it becomes pretty intuitive, pretty fast. Cause it's not unlike how you're already using AI.
- [23:12]
B
You can't use it on the phone though.
- [23:14]
A
Codex. No, I don't know about Claude. Actually. I'm brand new to their, like, premium plan and I haven't started using Claude code. I have the app, but I haven't like, even logged into my premium plan through the iPhone app yet, so.
- [23:25]
B
Got it.
- [23:25]
A
Everyone was talking about, like, all these MCP connectors and integrations and stuff. I'm like, screw it. Just tap into the API. It's way more robust because all the different connectors, like, there's a connector to high level and Codex. I was like, oh, great. Like, they already built the integration, but it's so weak. It's like a read. I can pull contacts and add contacts. I'm like, that's dumb. I want to, like, I want to say Codex. Take this email, turn it into an HTML template, and send it to everybody in my database with this tag. And it can do that. I've tested it already. I made a test tag, put me on it and I'm like, hey, send an email. The HTML, it was kind of a funky looking email. It was well designed but it wasn't the template or style that I wanted. So I'm still working on that. But I'm now using it. Like I probably won't be logging in. I'm logging into High Level less now and I'm just having my agent do everything through the API. This is the future. Salesforce itself said, hey, we're developing a whole platform so agents can log in and deal with Salesforce so you never have to log in again. They're calling it Headless Salesforce. This is an update from a couple of weeks ago.
- [24:26]
B
Wow.
- [24:26]
A
The problem with them is they make money per user. So I'm like, are they going to have to charge per agent or per call or per interaction? Now the big guys are going to struggle with this High level CRM or CRMs, like high level are going to like really take off because they have a fixed rate and there's no. They don't make money on users, they make money off. They make money on a particular things that are like usage based, like text messaging and emails. So they're already in a prime position to like destroy everybody. Because you don't want to have to code your own CRM and like stuff. It's better to just take that out of the box and then hook your agent into it to use it as a base tool and then code your own custom functions around it. In fact, it's a good segue into High Level, which is the sponsor of the show. Thank you so much, High Level. If you haven't checked it out, now's a good time to do it because I'm literally just having my agents run it more and more and it's already a solid base to work off of. So you should go to dances.com highlevel check it out, get an extended trial for 30 days and it blesses the show. Of course that's an affiliate link. So if you mean you can go and Google search it for a non affiliate link, but if you use my affiliate link, you know, you get a little longer trial. That's the best I could do. I'm sorry, that's all they'll give me. Check it out. It's an amazing tool. If you haven't checked it out yet, you really should go watch a few demos, check it out, go look at all the tools and how much you get if you're using HubSpot, I don't know why you pay for that anymore. It's ridiculous how much that costs.
- [25:41]
B
HubSpot. No.
- [25:42]
A
Yeah. I don't. I really don't understand. They're trying to figure out how to charge for AI usage and users right now because they're in between. But everybody's like, how do I even. Even budget for that? Exactly. I'm like, dude, they are screwed.
- [25:55]
B
Well, we didn't talk about the second thing on your point because we got so pulled into all the other tools. Chat, GPT Image Model 2.
- [26:03]
A
Huh. Have you tried it yet? It just came out.
- [26:04]
B
I mean, yeah, I've used it. I'm trying to figure out why is it better just even seeing the image for this, the thumbnail of this show, this episode, it's getting much harder.
- [26:13]
A
There's a few things that are better overall. It's better at like capturing your likeness if you upload a picture.
- [26:18]
B
That's what I was going to say.
- [26:19]
A
It's.
- [26:20]
B
It's harder.
- [26:20]
A
It's getting way closer to that. So that helps for people making thumbnails like the show does. But it's so good now. You could be like, render me a New York Times. I've seen this image on the Internet. It's like, render me an image of the New York Times. And all the stories are made up, but it's all cohesive and celebrating this theme. And it looks like someone just took a picture with their iPhone of a news of a printed newspaper. And all the text is perfect. The font, the layout, really image the way it looks on the. Yeah, I'll pull it up in a sec. Cause it's in the viral post for today. But like, it looks good. It's so good at text now that it can do full design work. It was good before, but dang, it's just a whole nother level when it comes to design and layout and text.
- [26:59]
B
So I'm in a meeting. We're doing like a strat op meeting where I'm like, we have all these focuses. We need to. As a lead team, we need to focus on the top three that are going to be the major producers, the highest ROI things, and we need to figure out what they are. So I go, great, let's spaghetti. Throw it to the wall. Brainstorm. We'll list out 10 things. I plotted on a bell curve on the whiteboard. I just kind of bell curve it. It's more to get what we should be working on in a certain quadrant. So I get it all plotted. I label the dots and all that stuff just with numbers. I've got the list on one side, bell curve graph on the other, just drawn. Take a picture with my phone and I go, this is perfect for ChatGPT. I go, Hey, I want you to take this image, make a list, and then make an image of the bell curve with the. With the points, the dots with the numbers in the right quadrant near the same spot on the line that I put it. Right. It could not do it.
- [27:49]
A
When did you do it?
- [27:50]
B
Two days ago.
- [27:50]
A
Try it again.
- [27:51]
B
It literally put the points and it would not, like, mark the right dots in the right location. I'm like, what the heck?
- [27:59]
A
Try it again. Put it on your phone right now and run it again. Are you serious?
- [28:03]
B
You think it'll do it?
- [28:04]
A
I think it can, yeah.
- [28:05]
B
I don't think so.
- [28:06]
A
Try it. Try it while I talk. You get it loaded up. The other thing that they fixed with this is before the image generator could only give you three pixel resolutions. And it was like horizontal, square, and vertical, right? Okay, like. And it wasn't even 16 by nine. It was like a four fives ratio, which is useful for, like, Facebook ads and some things, but sometimes you want very specific resolutions. But they caught it up. So you could just specify, hey, I need this 1080 by 1080. Hey, I need this 1800 by 450 pixels. And it'll give you that exact pixel dimension. I did it for the show this morning because I always needed as 1400 by 1400 square for the podcast thumbnail. And then YouTube needs a 201280 by 720. And now I could just specify that course because it's Codex. I can. Hey, every time we finish the thumbnails, give me these image assets and repurpose it into these sizes. And it did. So that's really helpful because oftentimes you're resizing and repurposing. So that's a cool update. It can give you actual. That was. That was Already true for Nano Banana, but I liked using ChatGPT to create the image. And then I would sometimes go to nano Banana to, like, resize it into all the different ad formats. So now you could just do it in ChatGPT. No reason to go to Nano Banana. I'm sure that'll change again when Google updates their.
- [29:20]
B
All.
- [29:20]
A
All their stuff in like, six weeks or five weeks whenever they're at conferences. I should have.
- [29:25]
B
I should have put this into Nana Banana to see if it could have.
- [29:29]
A
Yeah.
- [29:30]
B
Done it.
- [29:30]
A
Are you doing it? You're running it in chat.
- [29:33]
B
It's loading.
- [29:33]
A
Okay.
- [29:34]
B
I mean, it's pretty clear. It's not like chat wouldn't. I mean, shouldn't have a hard time reading this image and plotting it.
- [29:41]
A
See?
- [29:43]
B
Messed up again, bro.
- [29:45]
A
Okay. All right.
- [29:46]
B
It just can't.
- [29:47]
A
Still not perfect. But it's. It's gotten so much better. I swear. I'm using it for all my Facebook ads now. Let me show you. Let me show you what it did for me recently, because I'm using it for so many different Facebook ads, and now because of Codex, I can just generate them all at once.
- [30:00]
B
I'm going to try Gemini.
- [30:01]
A
Okay. Try Gemini.
- [30:02]
B
See how it does.
- [30:03]
A
Let me show you.
- [30:04]
B
I'm going to do the same prompt, same image.
- [30:06]
A
Look. Mother's Day promotional campaign.
- [30:09]
B
No, I know those kind of images
- [30:10]
A
look amazing, but look, all these. Look at all the packaging in it. Hold on. Look how many packages there are in this image. And these all look like the package.
- [30:19]
B
The base image. Did you have a base image?
- [30:21]
A
Yeah. What did I give it? Where is it? Here's the base image. Looks like junk. This is the base image. Oh.
- [30:27]
B
It's literally just like.
- [30:29]
A
It's just a Photoshop document of all the different product shots in one image.
- [30:33]
B
Just one image? Yeah.
- [30:35]
A
You can't even see. You can't even read all the packages. They're kind of like, slightly covered up because they're layered on top of each other. Yeah.
- [30:40]
B
That's terrible. Okay, so it took that.
- [30:42]
A
So hold on. I went through a few rounds, and then we get freaking this.
- [30:46]
B
Wow.
- [30:47]
A
And I'm telling you, all the colors, all the fonts on point, matching the style guide. I uploaded just an image of the style guide. Impressive.
- [30:55]
B
For those that can't see, it looks good.
- [30:59]
A
All the products are rearranged and all the logos on it are correct. That was like. I'm. I'm kind of sad because I've been running ads, but. And the products look good because Chat GPT was good before this, but the. The. The logo of the butterfly, logo of Trim Healthy Mama was always slightly different. And I could tell. I'm like, most people won't know, but I'm like, I know when my founder sees it, she's going to be like, why does the butterfly have look different on every single image?
- [31:26]
B
That's not good.
- [31:27]
A
Yeah, it's. It's subtle. It's subtle. But I was like, screw it. Time, efficiency, their ads, whatever. Now that's not a problem anymore.
- [31:34]
B
Okay, well, I just ran this image, this bell graph with Points on it.
- [31:38]
A
Well, what do we get?
- [31:39]
B
And it did it. It's funny, right? Because this is where it's like the bell graph is perfect. It's exactly. It even drew the line of the bell curve that I drew is even kind of like not perfect. Like it made it curved a little bit. But the points are exactly where I put it, whether they're close together or far apart. But the list, the design part of it, the text is terrible. So Chat killed it on the text part. Gemini killed it on the graph part. Oh, what are you gonna do?
- [32:10]
A
Just can't win. Can't win. And lastly, in the news, there's actually like a mountain of AI news just over the last couple of weeks, and I can only brush up over it. But, like, it's worth mentioning that, like, it's like a freaking model race. Like they're all the models are updating all the time. I feel like we just got 5.4, like not long ago. Like, like you can't even count it in months. We got ChatGPT 5.4 within weeks, maybe, maybe less than eight weeks ago. And 5.5 dropped yesterday. And you're like, my gosh. But of course, Claude's updating all the time too. So I think we're up to like Claude Opus 4. Seven now. You're like, mom and I, oddly, has been pretty quiet. I think Google saving all their best stuff for their conference coming up or their. Their developer conference in like a month. It's at the end of May. Okay, so. But everybody else is just dropping now. Anthropic has been pushing an insane amount of updates for Claude. I don't talk about Claude as much because I haven't been a huge Claude user. I'm still a ChatGPT guy, but I will be talking more about Claude in the future because I have a license and I'll be using it to push a lot of code and websites for my company. I still have my Chat GPT personal account. I'm using both so much now. But like, the models are updating at a pretty fast rate and the models are just getting better. Not to the point where they're doing a big number jump from like ChatGPT 5 to 6 yet. Right. But they're the micro. The incremental updates are improving. Of course, they were improving a lot last year too, and just not saying it now. They're updating every single time. They're putting out press releases every single time. So there's that. But it is interesting. Things are just getting better and Better and better and better at such a faster rate.
- [33:45]
B
All right, go ahead, Gemini. I just to follow up on this image that I've talked about. Probably too much, but a couple of prompts later, you know, like, hey, just clean this up, remove that, make the list stronger and cleaner, and it's did it right away.
- [33:56]
A
You got it. There you go, Gemini.
- [33:58]
B
Gemini. Crushed that one.
- [33:59]
A
There you go. Nano Banana. It's good at locking it in. Certainly really good at editing images once they're done and keeping it not screwing up other parts of it while you're editing it. Right. I do like that about Nano Banana. So how about your everyday AI?
- [34:12]
B
Yeah, I was thinking about this. I mean, I use it for everything. Like literally, if I don't. If I don't know something, I'm working with a lawnmower blade, I buy a lawnmower blade, I forget which way it's supposed to be oriented. I'm just holding it out with my camera and I'm like, is it supposed to be up this way or up this way? It's like that way. I'm like, oh, great, thanks. Latch down the lawnmower blade. And I'm like, I don't know. Could it have been a quick Google search?
- [34:35]
A
Sure.
- [34:35]
B
But it's like I just getting real time feedback with video on my phone is so helpful. That was literally an everyday AI jam, you know?
- [34:43]
A
Yeah, yeah, man. Reminds me of that video we saw of the video. Like the video audio model is still based off of a model from two years ago. Right. So you got to be careful with the audio video model.
- [34:55]
B
I know. And it's not great. It's not like fantastic, but it was,
- [35:00]
A
it was amazing when it came out a year and a half, two years ago. Blew everyone's mind. Now it's like rad. It's like so bad. It's so bad because again, it's. It's not even the four zero that we left off with last year. It was the four four zero when we first launched it two years ago. So it's. It's like one of those things where it's sycophantic and it's like confidently wrong all the time. That video you sent me earlier with like, hey, I'm gonna test you real quick. It's like, okay, go ahead. I'm gonna run a mile. Can you count how long it takes me to run? It's like, sure, I'm timing you now. And then he goes and runs, or he doesn't even run. He like moves his face away from the video model, he's like, like three seconds. He's like, okay, I'm done. How long did it take me? It's like, you took seven minutes and 36 seconds. It was literally three seconds. You know, it's kind of like, how did you know that? Because it actually didn't even run. It took me three seconds. No, I clocked you at 7 minutes 36 seconds. You're like, yeah, that's the audio model, unfortunately. Hopefully they update that soon. But it is good. I just used it. Last night, one of our little electrical outlets in our garage started sparking. I guess Amy pulled the plug out of it and like part of the plug got left in there and it's like started going crazy. So she killed the power, of course, and I had to go and change it. And so I found another little. I went to Ace Hardware and bought the replacement part and I, you know, turned off the power. I have a little voltage checker to make sure. I don't want to electrocute myself. I know people are gonna freak out the fact I'm doing my own electrical. It's not that hard when it's just a socket. Okay, I did put it in and even a little indicator light came on, but nothing on that whole line worked in the garage. I was like, well, what the heck? So you go to ChatGPT. I'm like, did I buy the right thing? You take a picture of the box, you take a picture of the sockets. Like it checked a few different things. It's like, yeah, you probably, you just wired it wrong. I'm like, I swear I put it back in the exact places. Like, yeah, that doesn't matter. Look at the back. There's a line on a load, white. There should be a white and a black. A line and a load. You probably put the line and the load and the load and the line. I'm always like, oh, you're right, it does say that. And I'm like, yep, that was it. Thanks, Chat. I love how it can just explain it to you to get. Because you're competent, but only like you're missing a few pieces. And Chat just fills in the blanks.
- [37:02]
B
Right. What else you got? Well, I've been going through so many documents at work, some legal documents, so I've just been uploading those constantly. And I'm like, I don't have time to do this. Here's a, here's a better example. We did a survey that had open ended questions on it and it asked two different open ended questions. And there were Hundreds of responses. So I just uploaded all of those responses into chat and I said, I need you to summarize the top five things that people are prioritizing. What is the thing that is mentioned the most? Give me a top five down list for each question under each of those questions. So when I read the question that was asked, I can read exactly the top five things that were mentioned the most to see what people really want. Instead of me reading through or manually doing it and did it in a second, it's summarized and then I could ask questions based off that. That was incredibly helpful.
- [37:54]
A
Yeah, it's gotten so much better at dealing with surveys worth of data now. Before you really had to be careful and spot check it. Now it's like, no, it's, it's, it's got tools now and it knows how to use them and it's, it knows how to reason through it and give you the goods. What I love to do now is actually just like put it in Excel or Google Sheets and just have a conversation with Gemini on the side. You'd be like, hey, can you build a chart for this? Hey, can you calculate this and build it into a new column? Done. Done. Gemini in Google Sheets is amazing. It's one of my favorite things.
- [38:22]
B
Wow, that's cool.
- [38:23]
A
My last one is, I think one of my favorite things is using codecs and the images together to just create all the images all at once. Because it used to take so long to do it. Be like, hey, here's the product shot, here's all the things that I need. Just make it and done like that used to be one of the most time consuming things. I'm now considering more things that can do because I don't do a lot of HTML emails because they're so design intense and they're just like one and done. So I do more some image, mostly text. But now I'm considering, I'm like, okay, since it can do any image size now, can it give me an email HTML and space the buttons out so I could just slice it? So it's like if you click anywhere on the image, it goes somewhere. I'll probably be doing more of those now with email and I'll probably shoot it. Might be able to like just code it up for me and send it straight now just like, okay, can you tap into mailchimp's API so I don't even have to log into mailchimp, just like send it, call it good.
- [39:09]
B
Wow, that's.
- [39:11]
A
I'm getting close to just being able to Send newsletters really fast now. Promos really fast now. And I'm probably just going to be solely working through Codex to operate with a lot of these tools. Some of the tools don't have API connections though. Like there's an ad manager and high level, but I can't, there's no API to deal with it. So I'm like, crap. Maybe I just tap into Facebook's API directly and manage the ads that way. I don't know. There you go, we'll see. Facebook's API is a pain to deal with. So I'm like, kind of like, nah, I don't know if I want to do that, but we'll see. Facebook actually hates me right now, but that's another podcast for another. That's another whole podcast. He won't let me in. And lastly, viral post is. I'm just going to share it because these are very typical. Every time there's a new AI tool, there's this account that I follow called Menchoi and he pretty much does like, everything's going to change because of this new tool. Here's a roundup of what people are doing. 10 wild examples. It's very normal post for him, but it is a good one because it's just fun to see like what other people are doing with the images specifically. So in this post what people are getting really creative with is they're creating like one image that's like a scene by scene and then putting it into something like Seed Dance and be like, play, play this whole episode out. Wow. It's literally like a really perfect looking storyboard with very detailed shots and there's like nine shots in a row layered in one image. And then Seed Dance can take it and then play those back to back to make a full little 10 second clip with all the shots in it. And it's like a whole anime looking thing. It's. It's beautiful. It's like, dang. I mean look at the animation and it looks freaking good.
- [40:41]
B
Yeah, it does look good.
- [40:44]
A
People are turning it into like, because it's so much better at images, then people are using it to create like images with Chat GPT and then running it through Seed Dance to like bring to life and bring audio and stuff to it. Here's like a fight scene where someone created a video game looking thing. People are doing whole like brand boards with ChatGPT's Image Image Generator. Here's like a video game layout of all the assets. Here's like a UI spread of like a UI interface. It Looks like a pregnancy tracking app of something. But it looks good, right? That's all just one image with all the assets in it. Wow. Let's see.
- [41:17]
B
Tragedy and Seed Dance is the Seed
- [41:20]
A
Dance is the video. That's the Chinese video creator that remember with a few episodes maybe a month ago we talked about it like creating a video in Hollywood or something like that. Crazy infographics like look at the amount of text. It's nuts. All the same.
- [41:32]
B
Wow.
- [41:33]
A
Like this is actual design level work from One Image magazine spread here we're looking at. That's a different level of image creation now. So I'm excited to see where it goes.
- [41:43]
B
Look at that.
- [41:44]
A
So things are speeding up. Codecs, clog code. You need to get into it. Get into it. This is the thing. And I'm going to be documenting it more and more on this show. In fact, a lot of the future episodes are going to be specific to Claude Code and Codex. Because ChatGPT is going away. These things are going to become the norm. I'd be very surprised if they didn't start just changing ChatGPT to operate like Codex and then eventually they merge back. I think that's where they're going is it'll just merge into one super app and people won't even realize they're that it's operating that way. They'll just notice that it got better and better and they could delegate to it and it can manage their files and all that stuff. Then we'll finally have it on our phones. So right now all the files are my computer. I'm sure there's a lot of people listening to that's like, well, you can commit it to a GitHub repo and that's like the next level that I'm going to figure out with my team is you can share all these files so that your team has access to the same files and documents that you do. So if you update here's. Here's the newsletter publishing process and you update the process with your ChatGPT. Your coworker's ChatGPT has access to those same documents.
- [42:43]
B
Really?
- [42:43]
A
You can just run the play? Yeah, that's. I've only heard of a few teams doing it like that, but is that. That is the future. And I'm. What I'm going to be doing with my team probably within the next two months, I will have this operational with my. My small team and I will build the processes and they will run the processes. That's probably how I will go. I'll probably make it so they can read and they can't write to it or something like that. For a few of the docs I don't know how that's the thing that I have to navigate is how do you create it so someone can't screw up something? You know, if you create all this time building out like this is the step by step process and some intern gets in there, it's kind of like no, rewrite the process to be like this. Oh, forget all that. And it's like all gone. I'm like, yeah, I mean that's why in GitHub you keep revisions of everything and you can see who changed what specifically. So like that's a code based problem too. And they've. The code people have already figured out how to manage integrating things or if you both change the doc and it gets pushed back up how to moderate between the changes. All that's been solved between code projects because that's how developers work collaborate on the same code base. It's all been solved but now we just have to figure out how to train non coders to use the same processes. So that's kind of like a little glimpse of where Codex is the future. But like learning how to do collaborative Codex environments is the next step. Level up from there. As if I didn't give you enough to think about already.
- [43:56]
B
Yeah. Wow, we've been inundated.
- [43:59]
A
So welcome to the future. This is what's coming. If you get into it now. You were still really early but a lot of people are digging into this. I didn't invent this. It just took me a long time to actually connect it of why this is useful for marketers. And that's it. So that's the pod.
- [44:14]
B
That's the pod Codex.