654: AI Does 80% Of My Job Now, But Here’s What It Sucks At

654: AI Does 80% Of My Job Now, But Here’s What It Sucks At Speaker 1: Welcome back to the podcast, the show where I cover all the latest strategies and current events related to e-commerce and online business. In this episode, Toni and I get into a deep discussion over how much of our business we've actually handed over to AI. We talk about what's working, the tasks we're still not ready to trust it with, and one moment in particular where AI completely missed something that a human caught in minutes. You'll know exactly where to draw the line between what AI should be doing for you and what still needs your own two eyes on. But before we begin, I just want to take a second to mention that I have a free e-commerce community that I'm incredibly proud of and would love for you to be a part of. It is a place where real sellers come together to share wins, troubleshoot problems, and support each other through the ups and downs of building an online business. You can join completely free at MyWifeQuitHerJob.com slash community, and I would love to see you there. That's MyWifeQuitHerJob.com slash community. Now on to the show. Welcome back to The My Wife Quit Her Job Podcast. Today we're going to be talking about humans and AI. We've just been using AI so much now, which has gotten me thinking, what do we need an actual person for these days and what tasks are better suited for AI? Speaker 2: I guess you want humans to babysit. Speaker 1: This era right now that we're living in is like my dream. Speaker 2: I know. It's like the Jetsons. Speaker 1: I've always wanted this because my philosophy has always been try to replace as much human capital with code as you can. That's why I always coded up all my own stuff. I don't like having a lot of employees and whatnot, and now we're actually getting in this era where AI can do a lot of stuff that a human can. Speaker 2: How long until you replace your workers at Bumblebee? I'm just kidding. Speaker 1: The robots, once those come out, I'm going to be the first in line. It's not going to be for a couple of years, I'm sure. Speaker 2: Actually, let's talk about that because I know that you, I don't know if this is true AI, but you have done a lot to streamline the embroidery process. Speaker 1: Yeah, but I mean, that's actually has nothing to do with AI. Speaker 2: Oh, I thought you had some stuff that you had just built that was like making that even better. Speaker 1: So I am working on it, although I'm not sure exactly how to do it. But okay, let me just walk through the problem first. So yeah, we get a lot of personalized orders. Actually, the bulk of our orders these days are personalized. And so that involves taking something from the website, And then turning it into this machine file that the machine can understand and then it gets sent to the machine and someone has to load the handkerchief or whatever and then it gets stitched. So I automated that a long time ago. But the problem is that it just takes what's on the website and then plops it into the program. Sometimes like the letters aren't aligned correctly. Like you know when you use different fonts and whatnot, like certain letters, Don't aren't spaced properly and so someone always has to go in there and look at everything and Then move things around a little bit for certain fonts. Okay, and that's the part that I want to automate with AI. Speaker 2: Mm-hmm. Speaker 1: And so it involves actually having the AI control the software unless it can understand the format that it's making and allow it to make the minute adjustments. Like that would save a ton of time. Speaker 2: Yeah. Speaker 1: And heartache. Speaker 2: But you're not there yet. Speaker 1: Not there yet, but that hasn't been my highest priority task. Speaker 2: Yeah, for sure. So I will say the biggest I would say the biggest impact to my life, time-wise, is having AI pull and analyze the, I wanna say thousands of reports that I have to pull, but it's probably hundreds. I have, right now, we were joking, I'm pulling a bunch of Klaviyo reports right now and analyzing stuff for the end of the year. And we're talking. This would be something that I would have to do all by myself. But I would have to log into Klaviyo, Go into campaigns, pull the exports, put them in a spreadsheet. I sort the spreadsheet, remove what I don't need, and then click through. Sometimes I don't even think, I don't know if Klaviyo exports with links anymore, but I would have to then go to those, analyze the text in the emails, right? Like all of that. So in theory, I'm analyzing two months right now. In theory, those two months of analyzing the emails from the past like three years would take me, I would say three hours. Realistically, to like do a good job, right? To go in there, figure out what's working, flag anything. Also, you know, I don't have to, like AI finds things that I don't, right? And we've talked about this. You were talking about finding a piece of code on your site. You know, so I think that right there, like you should not be pulling your own reports ever. Speaker 1: Dude, Klaviyo, the way I'm using it for Klaviyo is similar to yours. And Klaviyo, I mean not Klaviyo, Claude just flat out told me, do not send any more of these types of campaigns. They just do not convert. Speaker 2: Yes, yes. I always like it too because it's funny because I feel like everyone's AI probably speaks a little bit differently based on the interactions, right? So I get a much nicer Like, I don't recommend. This seemed to perform poorly. But yeah, I do that with all of our sales, right? So the post-mortem on a sale, once again, I'm pulling, you know, for like a huge, something like a Black Friday, right? Or a big, like a lot of companies have like seasonal sales based on what they sell. You could have 30 emails, right, that you have to analyze and it can tell you in a, I mean, sometimes it takes a little longer for it to pull the reports, but like while you're doing other things, it can prepare something for you that It takes you zero time and then you can just look at basically the answers. It's the answer key. Speaker 1: All right. So we already talked about how we're using AI. Like when does a human come into play? So I'm not talking about you, obviously. Speaker 2: Yeah. Speaker 1: Right. Like what? Let's say you have it automating your Klaviyo. Then when would when would a human have to step in? Speaker 2: So here's where I think the humans are not replaceable yet. So let's just say we get that campaign where Claude says, never send this again, right? But then you look at it and you're like, you know what? I sent that on a Saturday night. Right, or I sent this right after something else, or this was sent, like there are nuances that I don't think that AI is like familiar enough, and it probably will get there, right? But I feel like today, or you can go back and look at the copy and just, I think if you spend enough time in something, you know, like you'll see something, you're like, oh, okay, I didn't realize that I did that. And I think that's the part where I believe this is like my personal belief that the best AI comes when it starts with a human, right? Like the best copy comes from AI when it starts with input that you wrote, right? And it's just, it's just iterating on what you did. I think the best AI for analytics is like giving you all the information that you could possibly absorb. And then you looking at the nuances in there and figuring out like, you know what, you're right, this isn't good or you know what, this probably didn't do well because we were out of stock of this one level or this one product or we promoted it two months ago and it did really well. There's probably fatigue there and I didn't think about that, right? So I feel like that's where then I come in and I can look at that and go, okay, yes, this is true. We should never send this again or maybe we try this again but in a different way. Speaker 1: So you know what, since we're talking about this, one thing I did recently was I have one project now that has access to the ads, Klaviyo, and the database to my store. So it can access all that information that you're talking about. Speaker 2: Yeah. Speaker 1: I just don't know what to do with it yet. Speaker 2: So that's the other thing I find is sometimes, I found this, this happened to me on Saturday. I had all this information and I just didn't know what to do. I was like, this is all great, but. Speaker 1: Like they can all talk to each other now, but what I wanted to get to the point is like, okay, check out what we've sold recently in our inventory levels and whatnot, and then just come up with something. Speaker 2: Yeah. I feel like it's the same problem that happens in real life. One of the problems that I've seen with one of my clients is that I think inventory is hard for every company, but what was happening in the past was typically when you launch a new product and you do a big promotion around the launch, the sales spike because all of your ad budget is going there. You're putting a lot of email resources, probably SMS, lots of organic social. And then once the launch is over, it starts to like taper into probably the normal sales. Well, when you pulled like a six month inventory report, right? It's like, oh, we're selling 8,000 of this. It's like, well, really, if you take out the 4,000 that sold at launch over six months, we're actually selling like 1000 a month, right? But then inventory was ordered based on You know, the bigger number, right, because that wasn't factored in. And I feel like that can happen with AI. There's like, if there's one piece of information that it doesn't have, right, like this was a launch, then it's going to give you bad information, or it's going to give you a bad recommendation. And I feel like that's where the human comes in, where you're like, you have to tell it to know this, you have to tell it to analyze this or factor that in. And so I do think that area, that's an area where humans still matter, because Some of those nuances can't be picked up just from a report. Speaker 1: I just wanted to take a moment to tell you about a free resource that I offer on my website that you may not be aware of. If you are interested in starting your own online store, I put together a comprehensive six-day mini course on how to get started in e-commerce that you should all check out. It contains both video and text-based tutorials that go over the entire process of finding products to sell all the way to getting your first sales online. Now, this course is free and can be attained at MyWifeQuitHerJob.com slash free. Just sign up right there on the front page via email and I'll send you the course right away. Once again, that's MyWifeQuitHerJob.com slash free. Now, back to the show. I think like in theory though, AI should be able to solve those problems because you can look at all the orders and what happened and all the sales that you've run. Maybe this is a job for Fable. Whenever I do a project like this, I run out of credits pretty quickly. Speaker 2: A job for Fable. Speaker 1: Right. Exactly. Actually, you know, the other day I tried to use Fable for something and it refused to do it. Speaker 2: I had that happen too, but mine was totally different. What was yours? Speaker 1: Mine was I wanted to look for security issues and it refused to do that. Speaker 2: Okay. Mine was very dumb, but this actually made me really happy. I was watching my grandkids and we were out at an ice cream place and we were just kind of messing around. I was like, give me your wildest idea and let's make a picture of it, right? Because I like chat's image generator. So I opened the app and they were like, I want a flying cow with a tutu and it's farting rainbows or whatever, right? And so we're making these really fanciful graphics on chat, which I was like, this is very fun. Well, then we had the idea that it would be funny to put like one of the kids heads on like a lizard or frog, like make them part of the character and chat. So I took a photo of them like staying there in the ice cream shop and then I was like, take this head and put it on the cow's body, whatever. And chat's like, no, I will not do that with minors. Speaker 1: Oh, interesting. Speaker 2: And I was like, oh. First I was really irritated because I was like, come on, we're just trying to have fun here. We want to put our face on a lizard. And then I was like, oh, that's actually really good that chat identified that it was a photo of a minor and wouldn't do anything with the photo. So anyway, I was like, wow. A plus for AI for not contributing to the horrors of the world. Speaker 1: This is like a completely random story also, but you know how when you like you go on a roller coaster or something like that, they take a photo of you and then you can buy it, but there's this watermark in there. Speaker 2: Yes. Speaker 1: So now like with AI, you just say, you can't say remove the watermark. Speaker 2: Okay. Speaker 1: They remove this text and name this specific text and the lines and it does it. And so I think like those, no one's going to make any more money off of those photos. Speaker 2: No one's going to pay $80 for the picture of them on Splash Mountain. Speaker 1: In theory, I think AI can do all of these things. It just needs to train itself to do it. Anyway, back to the human topic. One thing for me that I've just noticed is for coding, when there's a right or wrong answer or math, AI is fantastic. There's nothing that can beat it when there's a definite right or wrong answer. Where it gets a little iffy is when the answer is ambiguous. That includes writing, anything creative and whatnot. I've been using Higgsfield a lot for all of my stuff actually, for images, for the website, for videos and whatnot. Someone always has to look at it for very subtle errors. Speaker 2: So we there was a conversation that i had last week and one of the and i will say there are a lot of people and i would say with good reason that are very concerned about like a i taking their jobs right because they i does. I would say for all the jobs that it takes, it probably gives new opportunities too, but that's a whole other conversation. But one of the arguments that I heard was that like, if you're using AI for image generation, and I'm not a graphic designer, like you don't want me designing your images. So I don't know how true or not this is, but I feel like there's got to be a workaround. That like the images that are generated are so low quality, you can't edit them, you can't put them in like a Photoshop or a Canva and make adjustments, and they're not even high enough resolution, like it's very obvious. And I was like, I mean, I've seen images that I know are AI, but I also have made images that nobody has caught are AI because I've just passed them through and tried to see if anyone would catch them and no one has. So what do you think about that? Speaker 1: So all those answers are false. You can generate 4k images in Nano Banana 2 and Nano Banana Pro. So the high-res thing is out. So that argument is flawed. And in terms of not being able to edit it, that's actually flawed too because Canva has the magic tool or whatever. The magic tool automatically puts things on layers so you can move them around. I mean, it's not perfect, obviously, but if I have like a YouTube thumbnail with text, my face, and like graphic, it puts those on layers and I can resize them in whatever I want. Speaker 2: Yeah. Speaker 1: So, you know, that's that argument. Speaker 2: Yeah. That's kind of what I thought, but I didn't have enough knowledge. Speaker 1: It's not perfect, but yeah, it works. Speaker 2: But I also think that like, so this is where, to me, if I was a graphic designer, right, or a creative whatever, I would be using AI to streamline my workflow so that I could take more clients on and make more money, right? That would be my strategy, right? Because I will say that if I tell AI to make an image, right, and I give it a color palette and I give it fonts, like I give it a lot of direction, right? A designer could probably still make a better image because they know that they want the text this distance or that distance and they want that the person is, you know, like, like kind of like when Dana Michelle gives tutorials on, you know, making videos and she's like, the light is coming over the left shoulder into the right. Like all that's like, I don't know that. Right. I'm not a photographer. Speaker 1: That probably started with an AI, right? Speaker 2: Probably, but I think that a graphic designer could make far better images with AI than I ever could. Speaker 1: Sure. Speaker 2: Right? Speaker 1: Yeah. By that same philosophy, someone wanting to write code, I would probably do a better job than someone who doesn't know anything about code because I know what questions to ask. I think that's the hardest part in learning any of this stuff. Speaker 2: So I think with graphics, you just need to be very like fine tooth comb, right? Like once graphics are generated, when my grandkids and I were making the crazy images, I noticed that the cow in one of the images had like a fifth leg, right? Speaker 1: Like. Speaker 2: Not really. It was like not really a full leg, but I was like, that doesn't quite look right. You know, so it's like, I think you still need to be very like fine tooth comb when you do that. And especially with, I've noticed errors in like spelling or not spelling, but maybe like the, I don't, I guess it is spelling. Yeah. It's like how the word is spelled, but it could be spelled different ways. It's probably, it probably got put in as an error, but like, I think you just need to be like very careful. So I think a human does need to, I guess, edit. Speaker 1: I was generating a video for a pizza cutting board. You know, a demo for it. An AI generated video. And it was perfect up until the very end. He had it sliced and then he displays the cutting board and the pizza like vertical. Speaker 2: And the pizza's still on it. Speaker 1: And the pizza's still like perfectly on it. It doesn't move. Like stuff like that. Speaker 2: So yeah, I think there's still a human component on that. Speaker 1: I mean, humans are still needed, but what is your philosophy on this as a small business owner? Are you going to just rely on AI and hire less, or do you think that having AI will make you want to hire more people because you'll have a lot more work? Speaker 2: I think the initial is higher, less, right? See, for me, I feel like I have an answer three, answer C, is that for me, in what I do, I feel like what AI has done is really freed up my time to think about other things. That's dangerous. Speaker 1: What are you thinking about? Speaker 2: I mean, you know, get my hair done, get my nails done. No, like it's given me time to like, so one of the things that I think we see with small business owners, right, is that they realistically cannot afford to hire more than like one or two maybe part-time people, right, because they're still doing everything. For how I see it right now, it gives a small business owner... I can't tell you how many people I talk to that are doing successful businesses that are like, I haven't looked at my email flows in a year. Speaker 1: That was me, actually, until I used Claude. Speaker 2: I haven't done a deep dive into this. One of the things that I didn't have time to do personally for one of my clients was do a deep dive into YouTube. She's been making YouTube content and like, I mean, it's going well. It's not like, it's not crashing and burning, right? But because it was going well, that was like back burner, right? It's like, we've got so many other fires, like that is like, just keep cranking out the content. Well, actually, this is a great example. Two weeks ago, I was like, oh, I don't have to manually do all this crap. I can have AI at least get me, you know, 90% of the way there. And I will say like I haven't found a good, and we did a podcast about this a couple weeks ago, I haven't found a great like fully automated, so there was still work involved on my part. But as opposed to being like two days worth of work, two and a half days worth of work, it was probably two hours worth of work, right? Analyzed a hundred videos. And, you know, came to some conclusions and one of the recommendations that, you know, kind of came from everything was I needed to update titles and thumbnails on some videos. Now, I know the people that are like really great YouTubers do this within the first like 12 hours, right? They're like, I just don't have time for that. I went and updated three old titles and thumbnails on videos and one of them immediately started performing. This video is two months old, right? Interesting. Speaker 1: Was it already dead, that video, or was it still getting views? Speaker 2: It was pretty much one view a day. Then as soon as I switched it, I think that day it had 37 views. And the click through rate increased from like 1 to 3.7. Like it was like significant change, right? And so now I have this list of, I still, what I need to do now, this is the human part, is I need to create a system that this can now do this for me on a weekly basis, right? So that I'm not looking at a hundred videos. So if that was my business and I was making the video content, and I know we have a lot of people in profitable audience that are making video, like that's the core of their business, right? I would set up a process where this was happening on a weekly basis so that I could be updating and then I would have the time to do that because instead of taking me two days It would take me 45 minutes and also I mean I regenerated the graphics. I used Canva because the original graphics were already in there so it was pretty easy to update but like you could use an AI tool. I mean Canva has a lot of AI tools in it too. All of that and it gave me title suggestions. Instead of me pounding my head against the desk going, okay, give me a better title. I've got to think of something. All of that was presented to me so then I could make the changes. If I was a small business owner, I would start doing that in the buckets of my business whether it's inventory or Um, you know, titles and thumbnails or, you know, I'm like emails, right? I have, I, we talked about this a lot. I have like a hundred flows, right? That there's no possible way you can like stay on top of that. Well now I can because everything is getting monitored through AI and so I'm getting like, Hey, this is, you know, not performing like it was last month. Is that a seasonality thing or is it because, you know, whatever, we need to change something. So I think a small business owner, it like gives, it's like hiring two people. Speaker 1: It's like hiring 10 people. So, you know, I recently just redesigned Bumble Bee Linens. I actually just went live last night, and I still remember many years ago when I had to work with a human, a human designer, and it was great. The interaction was great with whoever I was working with at the time, but each change required a day. I'd have to wait an entire day. Well, I compressed all that. Last time it took four weeks, I want to say. I compressed that into just one weekend. And arguably it looked better because. It actually read through all my site. I gave it my brand story, what we're good at and everything, and it created the website design, including all the copy. The only thing that it needed from me was images, and so what I had to do is I had to generate me prompts for the images using my products as a reference, and then it generated an array of images, which I used on the site also. So once again, it replaced an entire design team. And would I ever hire like a designer again? I don't know. It'd have to be like very compelling. Speaker 2: So here's where I would hire a designer. I think if you, so, I don't know how to say this without offending people, but oh well, here we go. I think some people just have really bad style. Like, and I see them put stuff into AI and it comes out looking bad, right? But they think it looks good, right? We have critiqued some sites like this. Speaker 1: No, but, I know, but the AI didn't generate the whole site. Like, AI's got great taste, because you always have it say, hey, look at the best practices for website design and conversion and whatnot. Speaker 2: But when you have people say, I want to use this hex code as color. Speaker 1: Well, that's very specific though, right? Speaker 2: What I would want a designer to do is give me some sort of initial direction. You had all this stuff with Bumblebee. You had all this historical content to help train the AI. Someone who's building their website from zero, they can pull from other sites. They can say, oh, I want to look like Pottery Barn or Crate and Barrel or whatever. I've seen enough design briefs to go, I would like a designer to turn the car on. I can take it from there, but have someone turn the car on for me. Speaker 1: But that's the thing. I gave it a clean slate. I said, these are my customers, this is what they want. Speaker 2: That's not a clean slate. Speaker 1: Well, I mean, you have to know if you have a product. Speaker 2: But not new people. Speaker 1: Think about it. Speaker 2: People who are just starting out. I still would buy some sort of branding kit. To get started. But that's where I would end it. I wouldn't go from there. Speaker 1: So I just did this exercise for a YouTube video. So Claude Fable went and grabbed my customers. It went through all the forums, wedding forums, Reddit and everything, and came up with what people are looking for to formulate the strategy. And then it designed the whole brand kit. And then I had it actually design a mock-up of the website for the redesign. The design looked fantastic, by the way. The only reason I did not make that drastic change was because I was just worried about the conversion elements and the navigation. Even though the website was beautiful, I wasn't sure if it was going to be as usable. Fable does a ridiculous job. That's why I was thinking, for your use case, arguably Fable might do a better job than. Speaker 2: Maybe. I still would like the human element in that part of it, especially if I have no data. And I like the idea, like I actually was on Reddit, I think it was just yesterday, I feel like it was like a year ago, I think it was yesterday, pulling information, right, and getting data on things. So yeah, I would say that one to me is, that's, I don't know, I just. Speaker 1: I get it. You're a client and I just went through a website redesign, right? Of course, having a human there to talk to and talk ideas through, obviously much more valuable than an AI. Let's say the web redesign costs, I don't know how much it would cost, but let's say it was like $50,000, right? Versus. Speaker 2: Fable credits. No, I get it. Speaker 1: Fable credits, which would be like maybe a hundred bucks, let's say, to do this over the course of a weekend. Like, how do you justify that 50k? Speaker 2: Yeah. Speaker 1: Like, do you know going in that your conversion rate is going to 10x because of that 50k and make it worth it? Speaker 2: Yeah. Speaker 1: Or can AI take you like 90% of the way there? Speaker 2: Yeah, so here's an example where I feel like a human is better from the process of redesigning. And if you know this, and now everyone listening will know this, so you guys don't need a designer, but you know how when you ask AI to generate, I don't know, like marketing sentences, we use a lot of like, you know how websites will have a bar at the top where it has like, stop our fall sale, whatever. So whenever you ask AI to do that, it always puts a little emoji like somewhere in the sentence. Speaker 1: It's never done that for me. Speaker 2: Oh, I see it everywhere. I see it everywhere. I see it in emails. I see it on websites. Like I've never had it not do it for me. Speaker 1: That's so funny. Speaker 2: You probably put into Claude like two years ago. I hate emojis. Never have anything happy. Speaker 1: No, I've never done that. Anyway, go on. Speaker 2: For a bit of time, we were using emojis like in that little bar, right, as like kind of a catch your attention. Well, for all the ADA stuff, it's actually, it's like a ding, right? So it's really bad to have emojis In your copy for people who have websites read to them because and so it's basically a recommended practice not to do that, especially in those call out bars and anything that's like a heading tag, things like that. I would have never known that. I knew an email that actually it's not always great to have it because of the whole ADA stuff, but I never translated that thought into the website. The person we were working with was like, hey, this is actually not best practice. This is actually something that's really frowned on. I would have never known that had I not worked with somebody. Speaker 1: When I went through the process, it actually did a sweep. It said, let me just double check your website for ADA. I did not ask it to do that. And it automatically did the sweep. Speaker 2: How did it do it without asking? Speaker 1: It did not ask me because it actually changed my shade of teal. I don't know if you noticed that. Speaker 2: Well, yes, because Klaviyo will tell you too, like their little AI will tell you. Speaker 1: It said, I'm changing your shade of teal for ADA compliance. And then it said, would you like me to run a full ADA sweep? And I was like, sure, why not? And it did that and it would have caught that. Speaker 2: See, it hasn't caught it in mind because I've had it build other things where it still throws those dumb emojis in there. Speaker 1: No, no, no, no, but you have to give it access to the whole project. So I was doing all this in Linux where it had access to every source file and everything. And then it had, what? What's so funny? Speaker 2: Okay, I'm going to stop you right there. Speaker 1: Okay, why are you stopping me right there? It's the same thing as having it on Windows versus I just happen to do it on the server. Speaker 2: But for a lot of people, they were just like, nope, I need to hire somebody. Speaker 1: Okay, well, let's say you're doing it on your Mac. Same thing, right? Speaker 2: No, I'm just saying in general, that's very overwhelming. And you get this in the class. Basically, every talk's on AI right now. But you get to the point where people are like, nope, Steve, there's no way that I'd be able to do that. Speaker 1: I know certain words I should never use. Server is one of them. Speaker 2: Yes, a little bit of code is another one. Speaker 1: I should just use Mac or Windows. Speaker 2: Yes, it's in the computer. Speaker 1: There's a best practices website guide that incorporates all that. Maybe that's why it asked me because I had to generate a best practices guide first. Speaker 2: Who did that? Speaker 1: You did. Claude did. No, of course not. Speaker 2: No, but you had it generate a best practices guide. Speaker 1: Yeah, but that's common sense, right? I don't know all the best practices. Speaker 2: Oh my God, I almost called you Brian because I'm irritated. Speaker 1: No, no, no, but that's the first step I always do. And if anyone's listening here, that's the first step you should do. Do deep research and find the best practices for whatever you're trying to do and use that as a reference. Speaker 2: Okay, but think about this. We are teaching people how to do this. If you are just caveman discovering AI for the first time, you don't know that that's a step you should take. Speaker 1: Correct. And that's why if you're listening to this, that's the first step you should take. I found all that stuff. I'm just trying to play devil's advocate with you here, like the human. What else did they do that an AI probably would not have caught? Speaker 2: I was not super involved in the whole redesign side, but that was one that for me was like, that's so interesting because I do ask for best practices. I do all those things and I still get freaking emojis generated in everything, which is to the point where I've had to say, I've had to add that to my rules of don't do. So anyway, I get what you're saying. I think there are... Speaker 1: I mean, I can answer my own question after I hear yours. Speaker 2: No, no, no. Go answer your own question. Speaker 1: Okay. So you know how I spend all this time on my YouTube thumbnails? Speaker 2: Yes. Speaker 1: And I have AI rate them. And I usually have like 10 for every video that different variations. But then like the AI is wishy-washy, so I'll ask four different AI their opinions. Speaker 2: Yes, and then you send them to me. I'm very aware. Speaker 1: And then if I send the same AI the same images, the answers flip-flop from run to run, right? And so in the end, I can't trust their opinion because there's no right or wrong answer. It's not code. Speaker 2: Yeah. Speaker 1: And so that's when I always ask you, I ask Jen, and then I ask my My editor, What They Think. And then through that I come up with the final thumbnail that I'm going to use. So I think you can have AI do most of what the designer wants or executes, but then you always need that final human opinion for the subtle things. For Bumble Bee Linens, There was some language it used in the tagline that it thought was fantastic. It was like, oh no, no, no. This is like way better than anything you came up with. But then like I had to massage it along to something that I actually really liked and then Jen wanted to represent the company, which AI probably would not have figured that out. Speaker 2: Yeah. So that's a really great example because I had this happen this weekend. So I was working on short form ad copy, short form video. It's founder based, so it's all coming from the founder. She's directly looking at the camera. I needed like 20 videos. I have my whole thing and all the little rules and best practices and all the marketing briefs got uploaded. It already has all this data of all the content and whatever. And, um, about halfway through the videos, like the hooks just went from like good to like terrible. Right. And I was like, like, what's happening here? And it's like, Oh, you're right. You're right. I got off track. Right. And I have found that whenever you're trying to do anything of like mass, right. So 20 videos is 24 short form scripts. Like, you know, I ended up spending almost two hours on these scripts because every time I got to a point, then it would stop. At one point, I was like, we can't end every video with the same line. What are we doing? And I was like, we need to do X, Y, Z. And I gave it a lot of constructive feedback. And it was like, you're right. And so then it updated the close. And then when we went to iterate again, it went back to its old ways. So then everything was like, look at this, whatever the sentence was. I feel like you have to really be paying attention. It's kind of like a little kid. You can't just let it, trust that it's gonna do exactly what you tell it to do. But then it was interesting because I worked on this, and I got to the point too where I was like, you know when you work on something for a long time and you're like, everything sounds the same to me, I don't like any of it. So I was like, okay, I think we're at a good enough point and I passed it along to the copywriter. And she is a true copywriter, right? And I said, hey, can you look these over? Feel free to improve on the hooks. Now, I know she's using AI too, right, to like work. And she messages me and she's like, hey, is there a reason at video eight you went from this type of speaking to this type of speaking? Which I didn't catch, right? One, because I had been looking at this content for too long and two, because AI just switched, right? And I didn't notice. And I was like, nope, absolutely no reason. So I do think that to me, I mean, I had been putting that through AI and best practices and all the rules and everything over and over and over again, right? And then she looks at it for five minutes and is like, hey, is there a reason why we switched the way this is being presented? And I was like, nope, AI switched it and I just didn't catch it. Speaker 1: Yeah, I 100% agree with that use case. The way you mitigate that, though, is you have each script spawn its own sub-agent with a clean context. And then you have another agent that's like the retention rater. I have this for my YouTube scripts. I have a retention rater one, which goes through each script and then gives it a score based on the guidelines. I also have a fact-checking sub-agent. The key here, I think, is each agent needs its own clean context window so it doesn't get degraded over time. It takes a lot of tokens, though. I agree. You always need that final set of eyes. To look at stuff I also think and this is why I like math in school better than English by the way because math always has a correct answer Well, I you know what? Speaker 2: I I mean, I definitely love English But I understand what you're saying because I remember in college, right, like if you took a math test or, you know, whatever, it was like you either pass it or failed it. You got whatever score, like the answers were right or wrong. And sometimes, and I, I mean, I did really well in college, so I'm not, it's not like, you know, I failed an English paper, but like I can remember turning in work that I thought was like really good and that I had spent, I remember I did a paper, um, The paper itself was so good that my professor was like, I think you should consider getting a master's degree in this. And I was like, I'm not getting a master's degree in women's history. What job am I going to get with that? But I can remember doing the same amount of work on another paper in another class and getting a B plus. And I'm like, Well, the effort was the same, right? I thought the quality was equal. Obviously, I'm biased, but I had one paper that was basically so great that they thought I should go and get another degree in it and the other one was like, you did okay, right? I mean, I guess a B plus is a little better than okay, but you know what I mean? What are you talking about? Speaker 1: That's failure. Speaker 2: I know. Thank goodness my parents weren't Asian. I would have been grounded for the rest of my life. But no, like, I never liked that subjectiveness, right? Because it was subjectiveness on all these levels, right? Content, the professor, like, a math professor looks at a problem, it's right or wrong no matter what professor looks at that, you know? Versus, you know, so I do feel like with the content side, having a human, and this is where I think if you are an expert in something, right? Like this person that I'm talking about is, she is a copywriter. She's so good. Right, and she can just look at something and go, why don't we try flipping it, right? And you're like, yes, that's amazing. Absolutely. Kind of how I feel like I can look at emails and go, you know what, let's try one thing, right? Because I've spent so much time in email, and while AI can get you there at some point, I can actually get you there a lot quicker in certain ways. So I think if you're like an expert in an area and you're concerned about If you're concerned about your job or you're concerned about being replaced, you have some key core part that, at this point, AI cannot beat you because you have just too much knowledge. However, by the same token, I think the people that are like, absolutely no, AI is evil, don't use it, it's terrible for you, are just gonna get swept away because the people that are at your level that are accepting AI are going to crush you. with the amount of volume and content they can produce.