How Pete Rakozy Uses Vibe Coding Custom GPTs to Build Faster and Smarter

(00:05) Hey everyone, I'm David Burkwitz and welcome to another edition of AI Insiders by AI Marketers Guild by March Media. And I'm excited to be here with a lot of community members and also our guest this week, uh, Peter Ricozie from Exit Loop. And uh and I mean we were geeking out on some topics like uh like vibe coding and custom GPTs and I was just so excited uh getting to learn from Pete and hear about some of the work he's doing and he was kind enough to offer to share some of this uh with our fair community. So uh Pete, welcome.

### Guest intro: Exit Loop and operations background (00:48)

(00:48) Excited to share. Why don't you fill us in a little bit more on who you are since I'm not sure I could fully do you justice, but 

> > yeah. Um, so I'm Peter Rosi. I'm co-founder of Exit Loop. Exit Loop focuses on helping business owners get out of the day-to-day through AI automations and other trainings.

> 

### Career path and focus on “the plumbing” of marketing ops (01:09)

(01:09)  and my background, I've been in marketing and operations for about the last 15 years. have done I think the track everybody's done where they've been an account manager and then worked their way up through an agency and then I started my own agency run a couple of those failed many many times because running agency's hard and then discovered that where I really excel is operations like I always gravitated towards the marketing operations side of things not so much the glitzy and gl glamorous like ad creatives and all that

(01:39) fun stuff too or the strategy I just like building the the the the the plumbing of it all, right? And so that's just turned into what we do now at Exit Loop, which is in short what happens after leads generated. We help a lot of service businesses and local businesses fix that pipeline. And then now because AI is so awesome and cool, it's training them on how to use AI in their day-to-day and then building new operational processes that leverage AI to help them do more with less, profit more with less. So yeah, that's that's

(02:14) my career in a nutshell.

### Personal background and interests (02:14)

(02:14) And then personally, um I've got three kids, twins that are turning four and then a toddler who's two. And they keep me busy. They're they are so much fun um and hilarious. And I live here in just outside of the Salt Lake Valley in Utah. I haven't done any snowboarding or skiing this year cuz there is none out here.

### Hobbies: outdoors and Zelda (02:40)

(02:40) Greatest snow on earth. My eye. There's none. Um and and when I'm not nerding out with stuff, um I try to get outside and go camping, fishing, hiking, things like that, or play some video games. I'm a huge Legend of Zelda fan, so I'm really excited to see what happens this year with the 40th anniversary, the new movie, all that fun stuff.

### Transition to AI work and keeping it practical (03:01)

(03:01) So, hopefully it won't be a disappointing year. >> But, yeah. >> Awesome. Yeah, I I I Yeah, it's still have the uh the original NES and uh I mean to to me nothing beats it no matter how good Call of the Wild is and everything. Yeah. >> Yeah. Yeah, that's true. Um well well yeah I mean we'd love to hear more about uh you know what you're doing in the AI space and what we can learn from you.

### Demos, tools, and community Q&A format (03:28)

(03:28) >> Yeah. Yeah. Well, I've got a bunch of like little demos that I put together that just shows like here's practical things we've done for clients and I can share those and then I've got a bunch of links to tools that I've built to help myself um that I'm willing to share with everybody as well and they can they can use.

(03:48) So, we'll hopefully keep this as practical as possible, not death by a thousand slides and and go from there. >> Yeah. And for folks who are uh newer to this series, yeah, we welcome questions come through and and folks come whether you want to be on camera or just share stuff in the chat. Uh but then um yeah, by by all means, yeah, we can just uh keep this conversational.

### Goal: a simple loop, templates, case studies, and tools (04:13)

(04:13) >> Sweet. Awesome. Well, I'll walk you through this. Yeah, ask questions along the way. Um, what I'm hoping I can give you guys today is just a simple loop, some copy and paste templates. I'll walk you through some case studies that'll hopefully spark some ideas or new ways to do things.

### Problem: finding custom GPTs across accounts; Crewar tool (04:29)

(04:29) And then of course, I got some tools. So, uh, if you build custom GBTs like I do or gems or whatever, my biggest problem has always been where the heck is it? Like where's that specific GPT I built for that specific purpose? And so, um, one of the things that I've used, uh, Vibe coding for is to build my my like solve my own problems.

(04:50) And so, one of the tools I built, I call Crewar. Um, and it's a free tool. You guys can use it if you want. Um, the the site and link and everything are here. But essentially, it's an app that organizes all of your custom GBTs, right? But what I like the most is that you have this quick search feature that allows me to come in.

### Quick search and “where is that GPT?” workflow (05:10)

(05:10) I can be like, "All right, cool. I I know I need to create um this transcript AI playbook, right? Where the heck is that? What account is it in? And if you're like me, I've got multiple Chat GPD accounts across my own and clients and things like that. And so I never can find where they're at. So this is one of the cool tools that just scratched my own itch and um and it it saves me so much time in actually finding where these custom GPTs are and where I built them.

### “Stacks”: grouping GPTs by workflow (05:39)

(05:39) And I'm working on some other cool stuff inside of it when I've got free time, which is never um to build stacks and share like your stacks and things like that. Stacks or are what I would consider a grouping of GPTs that I use for a task, right? So like for example, if I'm building an a marketing um campaign or some marketing content for a client, usually that's five or six different GPTs that I'm using together and I'd put those together in a stack, right? or we do AI audits for a lot of our clients. Um, and I've got a suite of

(06:11) custom GPTs that I use to execute that and put that together.

### Question: one GPT with many prompts vs separate GPTs (06:11)

(06:11) >> And so, um, can I just ask one basic thing and into this is 

> > Yeah.

> 

> > is like at what point do you decide cuz like you can have a custom GBT that you can have like a number of default prompts for. Yep. and tell it to do different things like at what point like like when do you decide to have one GPT that works with a number of different prompts and all that versus like when you want to break it out into something entirely separate.

> 

### Rule of thumb: one job, one GPT (06:44)

(06:44) My my rule it with a custom GPT is um one job one GBT like one output essentially right so and the reason for that is I've found if I've built like a um a Facebook marketing bot for example right it'll be good at strategy but then when I ask it to do specifics like different type of post type or content types or things like that um I don't get as good as results as if I were to break it up into three different bots like one for Facebook one for Instagram were, you know, the individual channel specifics like that. I found that the more

(07:18) specific I get with the actual task or job to be done, the the higher quality outputs, less revisions, less hallucinations, things like that. And so that's that's just my default. Um, I think of it in terms of an org chart, too. Like if I were to build a team of people to do this, then I would have a social media manager and then each of them would be a channel manager, generally speaking, right? And so I try to mirror my my bot development structure accordingly.

### Prototype manually, automate later (07:48)

(07:48) >> Um and and then the other question I get sometimes on this is well why don't you just automate this all through like Naden or things like that and usually it's because um you at this stage I'm usually still prototyping out what works right and trying to figure out which which sequence of AI bots will help me get the outputs I want.

### From master prompt to Replit/Lovable (08:13)

(08:13) And then once it's like finite and defined, then we move it into an actual workflow um to output things. But yes, anyway, but this was simple. This took me maybe like an hour to put together. And it was one of those aha like this sucks. I hate having to find all this. I used um oh, let's see which one is it. I used this custom GPT that that I got from good old Perry Belchure uh to build out these actual Vibe code prompts.

### Vibe coding workflow: work the problem in Chat/Gemini first (08:45)

(08:45) And this is usually where I start when I do any sort of I coding things is start inside of chat or or Gemini and really work the problem, work what it is I want to do and help create a master prompt first and then take that master prop into replet or lovable um and or or your vi coder of choice, right? and have it start with that and then I find it's much more efficient uh at credit usage but I also end up with a lot less junk code and broken stuff and things that happen just through thousands of iterations that typically come out of

(09:21) vibe coding bit by bit right and so that's one one cool tool I don't want to leave you with quick quick and easy something for you to to think about like what else like what are things that you're you you're doing um and and this is the process that I follow, right? It's it's the in creating these what I would call growth growth assets.

### Building only the “one little thing” instead of buying full software (09:42)

(09:42) So, I build that for me. But often when we're working with clients, they're like, "We want to do this or we want to use this software, but we only want this thing out of the software. We don't want to spend $10,000 a month for, say, HubSpot or whatever because it's, you know, out of their price range.

(09:58) They only want this one little thing." >> And so, we'll say, "Oh, cool. Well, let's take that. Let's see if we can just extract that out." And that way you don't have to buy the whole software. You just build the part that you want um and build it the way you exactly that you want. And I'd say probably eight out of 10 times um you're able to do that using your vibe coding platform of choice which is awesome, right? And so starts with, you know, that that initial vibe master prompt and then we take it into the interface and build that out. I

### Replit as an all-in-one build + integrations platform (10:29)

(10:29) and then if needed, I'll connect the brain to it, which is usually um something I've worked out in the custom GPT. Uh the cool thing about like Replet's my favorite right now, but it's become the all-in-one like I don't have to go and daisy chain stuff from NN and or Zapier or all these. It handles most of the workflow stuff internally, and it's doing a pretty dang good job.

(10:51) Um, and it's got built-in integrations for most of the LLMs as well, so I don't have to actually pull in my own API if I don't want to. Um, and then I could build out the bots's brain inside of the the vibe coding tool instead of having to pull it in outward um from an external source and then iterate and deliver, right? Super super simple.

### Prompt-building GPT: live demo and “coding brief” style output (11:15)

(11:15) So, yeah. So yeah, like we're talking about one user, one job, one screen. That's that's really kind of my focus. And so um but yeah, this vibe coding GBT um which I'll share with you all super super super useful. Um you just tell it the idea that you want and say, "Hey, help me build an app." And I can say, "Great.

(11:38) Let's go ahead and work through it." And now it starts to give me like here's the the template you would use. I usually stop it and say, "No, I don't need that. That's okay." Um, do you guys want to go ahead and test this out now? Anybody have an app idea they want to throw out and say, "Help me build this prompt customizer to save variables.

(11:57) " Uh, let them fill them in. Cool. So then it builds out the the plain English prompts. And and there's a question if this uh if this custom GPT is specifically designed to use Reflet. 

> > It is it's specifically built for for Replet, but I've used it in Lovable as well and I've had great results with it.

> 

### Breaking down app requirements and success checks (12:40)

(12:40) Um because all it really does is it it breaks down in plain English essentially a brief of here's what we want to do.

Here's the deliverables. Here's the pages we want to have in it. Here's your rules and limits. Right? Safety and access. Here's the look and feel.

Um and then it it goes into more, right? Here's your success checks. Here's stretch ideas that we can include. And here's notes for the builder. Right? So it it chunks everything down as like almost like a coding brief, like if you were to sit down and build this from scratch with an actual coder.

### Replit design-first feature (13:08)

(13:08) Um what are most of the questions they would ask to help start to plan this out? So then we take this, we go over to Replet, maybe we drop it in, and bam, like it builds out our first iteration. Um Replet has rolled out this new design feature, which is cool because we can actually take this and have it build out the look and feel first before it builds out the brain and guts.

### Case study 1: quiz funnel redesign vs agency quote (13:31)

(13:31) Um, and that saves you some time as well in terms of if if that's most important to you is I want to make sure this looks a certain way before it functions a certain way, then great. So that that's cool. Um, another use case that we've had, let me see. Oh yeah. So So we had a client who they had a quizunnel and this is essentially what what it looked like.

(13:56) like they answered some a basic form. They had this form that they filled out and then it took them to a results page that essentially said, "Here's your results. Let us know if you want to if you want us to talk to you about it." Right? And then we're like, "All right, this is this is garbage.

(14:12) " So, we asked their their their web development team who managed their website, which I think was built on Web Flow, um if they could redesign it and make it much much better. And they agreed. They're like, "Yes, the user experience here is garbage. we need to make this way better. And they said um it would take them about 12 weeks to build out and I think it said they said it would cost about 80 $85,000 for them to to create all this stuff.

### Building a better quiz funnel in ~4 hours (14:38)

(14:38) And I was like, okay, well, they're a small company. They don't have a budget for that. And that's also a long time to wait. So I then said, well, let's just take your existing everything that you have in there. And I took all the screenshots. And this is before I had the VI coding. um GPT that I use now. Took all the screenshots, uh ran it into GPT, had it helped me build out a prompt, and then we built out this this uh awesome quiz funnel.

(15:07) Uh and I think I've got it here that we can look at. Yeah. And so I took their branding, I took their brand voice, I took all that type of stuff, and then it built out this beautiful page that used their images. We got all this type of stuff. And then they can go through and complete this assessment. And this looks way better.

(15:30) It's than before. It's on design, all that type of stuff. And then like a good quiz funnel should, it asks for their information. Um, and then uh it gives them a customized results page. And this whole thing took me probably about four hours to put together.

### Implementation details: CRM, webhooks, and admin dashboard (16:06)

(16:06) >> Um, and I think I did this while they were on vac like the the owner was on vacation and he came back and he was like, 

> > "This this is awesome. This is what this is really cool, right?" And he had just launched a book, too. Um, and so we're able to use that book to help field this information. And so it gives them this custom everything. Here's what their next level opportunities. Three of those four hours were spent on getting this thing to work because each one of these assets um changes color based off of an answer.

> 

(16:30) And Lovable is having the hardest time working with circles and coming up with this. Um, and then it links to their their calendar, but it also since you can connect web hooks and whatnot to to log, we're able to take all this information, push it into their CRM, which is go high level, and it then populated um their contact fields and set up all the nurture sequences, sent their, hey, here's what your results mean. Here's what you can do.

(16:59) Here's some quick wins. and and this has become a great asset for them um to use. But on top of that, we also built them an admin dashboard because one of the things that that Oh, shoot. Can I I don't have the login to that anymore.

### Making the wheel interactive; extracting existing code (18:01)

(18:01) >> Um, but the admin dashboard would show a list of all the the people who had taken the the exams or the the quiz and their results and then they could quickly pull those up and use those as talking points inside of their discovery calls when they' review this.

(17:35) Um, and it made things super super handy. And then of course you always have people are like, I didn't get my results. Can you resend it? And so we had to do some quality of life stuff like that where where click oneclick resends and and things like that. Um,

> > and and just like with with the wheel itself because like you know one of the things that I often get stuck on with vibe coding is the sameness u of a lot of the outputs of them.

> 

And so in that case was like that asset in their book and like they'd created that so it was a matter of adapting it and making it interactive.

> > Yeah. Exactly right.

> 

(18:25) like fortunately I lucked out and um they had this wheel on their website and the coding company that had built that um had all the the the code to create that already.

(18:25) So I just had to have Claude go and essentially comb that page and extract that code and figure out what it was that did it. And then Claude rebuilt it for me so that I can then take I just took that code snippet and gave it to Lovable and said, "Hey, like here's what you need to do because Lovable was not having it.

(18:46) It could not figure it out for the life of it." And so um so that's what I had to do and it and it worked worked great, right? And uh to do that.

### Capturing and routing data via webhooks and CRM integrations (19:08)

(19:08) So yeah, cool little asset, right? They they had a 

> > underperforming funnel. Four hours later, we got them something that was much more high performing, high converting, high quality, especially with their price point.

> 

(19:08) Like they they charge 100k to go do these branding audits, right? Where they do a deep dive, interview tons of people and figure out what that brand unlock is to to get people unstuck and to scale.

And so, yeah. Anyway, so that's that questions.

(19:08) Um, there was one specifically about like just like email capture. So then I guess you can tie that into whatever CRM they're using or

> > Yeah.

> 

(19:37) >> Yeah. >> Yeah. It's all done through web hooks. And so the web hook captures all the information and then you can push that information to where you want.

(19:52) And so love will just set up like I just said, hey, I need to integrate high level. And it's like, oh, we've already got an integration built for that.

(19:52) So let's go ahead and connect that. And it's like, here's the web hook URL you need. add that to your workflow and it actually walked me through what I needed to do inside of Gohigh level to set it up which was great um to to have that handholding which was nice.

### Case study 2: selling AI audits with custom GPT + Gamma (20:26)

(20:26) Mhm. >> Um and yeah, and it captured the email, captured all the information, uh captured all of their results as well and put that into to uh go high level because then we could use all those custom values for sorting and personalization and and any future campaigns they wanted to run to people

who, you know, scored red on this category of their brand wheel, right?

> > So cool. All right. So that's case study one, something that we did that was fun. Um, anyway, case study two. So AI audits, that's one of the things that we've done to to generate clients is a lot of people want to talk about AI, but they don't want to pay the high ticket required to implement it.

> 

### Interview transcripts → audit GPT → presentation and pitch (20:57)

(20:57) And so we um came up with a process to do um it all these audits. And so we'll sit down and we'll do interviews. We'll record those with Loom or not Loom with Zoom um or Google Meet. And then we take those transcripts and those transcripts I'll then run through um um this this not that one this vibe audit GPT that I built.

(21:18) Essentially, one of the things I love about custom GPTs is that you could you can create the amalgamation of some of the brightest minds in the world around a topic.

### Channeling ops thinkers; debating outputs (21:18)

(21:18) And so when it comes to like operations, um, like I can say, all right, you need to channel Peter Ducker or Eli Golgrat or, you know, any of these the these uh industry greats who have revolutionized operations and and theory and all that type of stuff, right? 

> > And I could put that into this custom GPT inside the system prompt and have it do that. And then I give it the actual

> 

(21:53) format that I want it to output.

(21:53) So, it takes and consumes all of these transcripts and we'll interview 10 to 15 employees. We'll drop it into here. It then gives me an output.

(21:53) Um, often times I'll debate with it and say, I don't know if that's right. What about this? And and that's kind of the cool part about about working with these custom GBTs as well is is it'll spit out an idea and then I'll be like, um, let's refine it.

### Presenting in Gamma and tailoring language to audience (22:17)

(22:17) And so it feels more collaborative than just throw something in and and take the output and and hope for the best. Um, but then it gives me the output I need and I can take that over and put it into gamma. Have you guys used gamma at all 

> > for creations?

> 

> > Yeah.

> 

(22:53) And so I'll throw it into gamma and then it'll it'll create um these audit presentations for me, right? And so I'll then take this and this is what I'll present on and say here's what the total impact will be if we fix these things based off of your audits, right? And here's what the

blueprint is going to look like. Here's what we diagnosed.

(22:53) Um, and this like and that's the great thing I love about AI too is I can change the verbiage to match who I'm talking to.

### Audit offer and conversion rate (23:22)

(23:22) So if I'm talking to a doctor, we can then use language like prescribe, diagnose, protocols, things like that, right? 

Um, and then pull in actual like quotes saying, "Hey, here's what Celestia said about this or here's what Cressela said about this or or things like that.

(23:48) " and then break down here's what you need to do. Here's your primary bottleneck. If you fix this and alleviate this bottleneck, then your throughput will increase in these parts of your business and that'll increase cash flow or time or whatever it is. And so this is what we present and we sell this for anywhere from $1,500 to $10,000 upfront um to build something like this and present it to them.

### Time savings and AI interviewer idea (24:10)

(24:10) And the value is in the clarity that they get. And and then this also helps break down the the impact of in action as well. So if you guys don't do these things, here's what your total monthly impact is going to be, right? And and then our offer is now we take this amount of money that you paid us to go in and do this deep dive, which we normally do anyway.

(24:36) Um if you were to skip that arch, that audit and just work through our 90-day core um process launch. uh then great, we'll credit it towards your next project with us.

(24:36) And our conversion rate, right, so far has been like 90% of those people have been like, "Yeah, heck yes, let's move forward." Uh the rest just disappeared into the void week of between Christmas and New Year's and haven't gotten back to us yet.

### Building the audit product quickly and iterating (25:01)

(25:01) But anyway, yeah, this whole process like normally would have taken me, I don't know, 10 20 hours to put together, but um excluding my interview time, which we're working on building out an AI interviewer that can actually do the interviews with everybody at the company instead of what my calendar allows. It takes me maybe two hours of my time to prepare for this audit presentation, which is great.

### Case study 3: OnMessage prototype and app platform landscape (25:46)

(25:46) Most of that is me just spent reading and reviewing this and really understanding the what we're putting together.

And then the rest is debating with my AI ops custom GBT to to be like I don't agree with what you're saying like fight for it like tell me why we should do that and and um making sure that we're in agreeance on what what that actual uh solution presentation should be. So cool.

### Building “OnMessage” prototype in ~1 week (26:28)

(26:28) So, this was something that, no joke, like I watched a webinar on somebody else who was doing AI audits and then we built this product in about an hour based off of that and then went and sold it and started generating money um through that and then just improved the system along the way as we've gone through.

(25:46) As you can see, we're on version four right now of of of the system prompt and tweaking it and improving it.

(25:46) Um, cool. So yeah, the AI auto engine, that's one of one of the things that super super easy to do, something you guys could steal, execute and run yourself as well.

### 16-month dev project → Lovable prototype (26:48)

(26:48) Um, another cool thing uh that we did with Vibe Coding before Vibe Coding got really big uh like it is now is that same client Backtory Branding um they had been working on this thing called um OnMessage and Ones Message essentially was their attempt at taking all these branding

assets that they capture when they're doing these interviews and distillation to try and figure out what what's the brand unlock.

and putting them into a software format that allows you to um create all of your content in one spot and make sure it's always on brand. Like it goes through all the filters, all that and and all that stuff.

(27:22) They've been working with um a development agency for about 16 months and had barely gotten to a point where they had a workable prototype.

(27:22) Um and then uh just for fun, I was like, I wonder if I can vibe code this. I wonder if I can build this whole thing inside of Lovable.

### Screenshot-driven spec and training the internal product owner (27:22)

(27:22) So, I went in and I took screenshots of everything, every single page that that they had built out and then I I had one of their employees who had been the product manager on the software and we talked through all the functionality and then dropped that into the level and then they within uh about

(27:57) a week had a working prototype that um they could actually demo to clients and show what was working.

(27:57) Um, and then I trained that employee on how to take it and run from there.

### Building full AI-powered software in Lovable (27:57)

(27:57) And they've built out like a full-on AI powered software essentially like it'll scrape your website and it'll scrape all your assets and everything and autofill all the stuff and then they've integrated AI into it so that if they need to create a blog post, they create any content, it all runs through the correct um, branding filters 

> > and creates things um that are awesome,

> 

(27:57) right? Like the feedback I've gotten from clients is I wouldn't change a single word of that or that's exactly how I would I would write that or things along those lines, right?

### Hosting and scaling concerns; security anecdote (28:23)

(28:23) >> Like because this this also comes up like Yeah, I I see some of where Gemini is going with this and and Claude tends to get pretty high marks.

I know from folks who are 

> > Oh, yeah. into coding.

> 

But then um but then like the hosting winds up being trickier, right?

I like like for right. So like for things like that then you still need to have typically like a another hosting service or like a VI coding app like a lovable or base 44 or something that will actually go and uh and yeah deploy it for you.

### Developer pivot: “big boy app” hardening (29:21)

(29:21) because 

> > Yeah. Yeah. Replet's got built-in hosting. Yeah. Lovable, they've paired with Superbase, right?

> 

> > And Superbase does the hosting for them and and so they've made it really both of those two have made it really easy to to launch.

> 

(29:43) Um just connect a domain and you're good to go.

(29:43) Uh but yeah, I I see that as an opportunity for developers now.

Like I've got buddy, he runs a company called PXP coding and he's had to pivot hard from doing what he was doing before which was websites and department of defense contracts and all these types of things to he's essentially like once you've viodated your app and now you need to make it a big boy app.

### Security + scalability limitations of vibe-coded apps (30:15)

(30:15) like he takes it from there and goes and fixes any problems, sets up the proper security, gets it ready to scale, makes updates to the codebase so that it can then grow into something that's more mature because that's kind of the the restraint or constraint we

have right now with most vibe coding apps is that um people still kind of have the opinion that um this isn't real software is kind of the the word I would say like this isn't something that you can scale to a million users, Okay.

(30:41) Um or 100 million users like you actually need to to take it and have um I'm not a coder so so I don't I don't know what I'm talking about.

(30:41) I know that like you have to take it and there's other things you have to do to actually make it secure and scalable and it can handle um the amount of users that are using it that type of stuff right which I don't know if Replet base 44 like gem like they can actually handle that type of stuff um out of the gate right now especially after I think there was there was a I don't if you guys remember the app it was like te or something like that that was a big app that was completely completely vibe coded and lovable but had no security in place.

(31:12) And so

so it uh uh all the user data, chats, everything like that got leaked as a result.

### Custom GPT builder to speed up system prompts (31:36)

(31:36) But anyway, but yeah, so that's another cool thing um that we built and and go from there.

### GPT builder walkthrough and client use (32:12)

(32:12) So, so yeah, those are here's the tools that I'm that I'll share with you.

Um, oh, and then I forgot about this one.

One of the things that that takes a while to or used to take a while to do is actually building your own custom GPTs, right?

So after building manually building my own custom GPTs, I built my own custom GPT builder and

um it essentially walks me through uh the process of creating uh a very strong system prompt and associated files to attach.

(32:31) And uh what I'll do is I'll just sit and I'll brain dump.

Here's all the things I need to do. Here's who I want to uh talk to. Here's what I who I want it to channel or here's the goal. Here's the purpose.

And then it helps you really narrow in on here's what you need to do or oh that's three different GVTs. Let's go ahead and and create three system prompts for you and and go from there.

### Audience question: voice mode and teaching clients to “talk to AI” (33:10)

(33:10) So it's super super easy, right?

> > Um to do that, you just come in, help me start from scratch, and then it'll start asking you questions to figure out what are the things we like it needs to know in order to help you start building this GPT from scratch, right?

> 

And so this has helped me save tons and tons of time, especially if when working with clients.

Often they'll be like, "We're stuck with this."

And I'm like, "That really isn't an automation. You just need a a custom GPT to help you out with that."

### Dictation tools and GPT personality settings (34:04)

(34:04) And so we'll take that that transcript.

I'll drop it in here from that call and this will help me get the initial version one of the system prompt and built out in about 15 minutes.

(33:10) And then I'll give it to them to start testing and and using right away.

And usually they're like, "Aha, this is amazing. This is this is super helpful tool. What other questions? Yeah. What other questions do we have from

> > Hey, this is Earl. Just want to say thank you so much for sharing all the different ways you've applied custom GPTs.

> 

It actually, you know, ties into the my approach to how I handle AI.

I have a question with um I guess the chat GBT 4 5.2.

Have you been using the speaking function?

because I think I've been finding a lot of success teaching clients how to talk to AI who are not comfortable typing or really uh self-conscious about the prompting prompt or context engineering

because I used um the voice chat like back and forth a ton um when it was 40 and then after 40 for whatever reason it seemed to get dumber and worse.

### Human review and debating the AI (35:24)

(35:24) >> I haven't I haven't used it since five um or five at all.

Um, mainly because I use so I'll use just the the the dictate function a lot and just talk to it or I'll use a tool like Whisper Flow

> > um

> 

> > to to brain dump everything into it because it's got a much longer um and faster transcription speed in my opinion

> 

> > and I'll just drop it in and paste that and go from there.

> 

(34:32) Yeah, I I don't type hardly in at all anymore and I I try to teach clients the same thing, like just talk to it.

Just that's the most natural way for us to actually engage with stuff, right, is just to talk.

(34:57) And so, um, yeah, the great thing about the custom GPTs, too, is that it's once it's already programmed on its job, like it helps eliminate a lot of the confusion or the the need for clean prompt structure, I guess, because it's it's kind of built in.

(34:57) And so, they can't just ramble and brain dump everything into it.

And then the AI will be like, "Okay, cool." Um, let's do this.

### Question: which tool is most “ready to go”? (37:00)

(37:00) >> I also, and I, I've also recommend if you haven't done this, changing your Chad TPT's personality to um, sarcastic

> > and and uh, it gives you it's much more to the point.

> 

It's dry, hilarious humor, which is my humor.

Um, and uh, yeah, it's much more concise and it's deliverable as well, I think, which is awesome.

> > As I'm watching you describe your process, one thing I'm not hearing you say is, and then I read it over and and bring human discernment to correct it or to make sure that if it's a presentation, I can stand behind it and present it to the client.

> 

Like, are you doing that?

### Answer: Replit reliability, Gemini for “how-to”, ChatGPT for strategy (37:33)

(37:33) >> Oh, yeah.

> > Yeah, definitely.

> 

Yeah, there's there's always uh um that's where most of my time is probably spent, right?

Generation is lightning, which is great, but most of my time is spent reviewing and debating with with the the AI like like because even when you try and change all the settings and everything to be very flat and not not pleasing, AI is very much a people pleaser and so it will like try and agree with me on everything.

(36:23) So, I find most of the time I have to be like, "All right, I disagree with you. Defend your stance." Like, debate with me on this type of stuff. That's where most of my review and feedback comes in.

(36:41) So,

> > I always gut check it because there'll be things it'll create.

> 

I was like, "Where did that come from? Like, we didn't talk about that at all.

Why are you throwing that in there?" And it'll be like, "Ah, you're right." Um, or if it's sarcastic like it'll be like it'll be like, "You gave that to me, not or whatever, right?" Like, it'll it'll just toy with me.

Um, but yeah, you definitely have to review everything.

It's not just spray and pray blind faith, right?

### Managing outputs and assets from GPT work (39:48)

(39:48) >> Okay. No. Cool.

Cool. Because that's my experience, too.

And I was like, wow.

So,

> > yeah,

> 

> > that that's yours.

> 

> > I I have a question.

> 

> > One one quick followup on that and then uh and then I won't steal your time.

> 

Um, but have you found for any of the tools you're using, Pete, like whether it's lovable, chat, GBT, Yeah. replet gamma any of them that the accuracy is like the closest to just being ready to go that you have to proof it the least.

### Question: tracking assets generated by GPTs (40:14)

(40:14) >> Oh yeah.

So replet I think is is the most dialed in for me right now in my experience.

Um I get the fewest errors.

I get the fewest breaks.

I get that type of stuff with with it.

It's still there.

Um and then with Gemini I I I have been I guess doing double prompting like I do everything in custom GPT or in custom GT or chat GPT but then I also do the same thing inside of Gemini and I compare the results and it's interesting I will find for strategy and business type of things.

(38:05) Chat GBT for whatever reason seems to be better.

Like it seems to be maybe it's because of its ability to access all of my chats and have kind of a wider memory on on what I'm doing.

Whereas Gemini is still very siloed in their conversational threads, right?

Um it it then takes a lot more things into consideration when debating with me on what strategies we should do.

(38:29) Um I find Gemini is most accurate when it comes to how to do stuff, right?

I'll be like, "Hey, walk me through this."

> > Or, "Can I help?

> 

> > Do you have any little like little band-aid things?"

> 

> > I don't. Sorry.

> 

Um, but

> > we can cut it.

> 

That will be

> > um but yeah, Gemini is definitely I think much more

> 

> > solid.

> 

I'm just losing and things like that.

### Audience member’s tool for saving chat links and summaries (41:03)

(41:03) >> Okay.

You showed us this tool you use in order to search for the custom GPTs that you created.

> > Uh-huh.

> 

So, what about the assets that are generated from the GPT? Do you have a way to manage all of those?

> > Um, you mean like the chats and things like that?

> 

> > Well, I mean, you create something and then you have all of these documents.

> 

Uh, how do you manage that?

Uh, I guess it depends on the assets itself, right?

So, if I create things, a lot of those things end up in Google Drive, right?

If they're text or copy or things like that, images as well, right?

All I mean, both platforms have image libraries that you can look at and scan and and grab those images regardless of chat, uh, which is nice.

(40:40) But, um, I I anything that I'm creating, I always dump it into one central location.

And so, we use ClickUp for all of our client management.

And so, and then we pair a a Google Drive folder with that.

And I'll just drop all put all those assets into a Google Drive, a Google Doc, um, and keep it organized there.

Or drop it into the task inside of ClickUp so that it's organized there and connected to the task.

So, that's how I'm keeping those things.

### Crewar auto-scraping GPT links and second brain reference (42:05)

(42:05) Are there specific types of assets you're you're

> > No, I'm just wondering because I I have a tool for that.

> 

So, I was wondering because that's a problem, right?

I mean, it you create

> > a lot of different assets.

> 

Some of them you want to keep, you know, I call it the endless scroll on the chat GPT.

> > Y

> 

> > uh after a while, how do you how do you keep track of anything?

> 

> > So, you know, we I have a tool that

> 

> > is hooked to a you know, you can create custom GPTs.

> 

Everything that you do in GPT or any application creates a unique link

> > and those links you can save.

> 

And I actually have a custom GPT that allows you to summarize the description.

For example, for my GPTs, I say, "Give me a summary and list all of the questions I asked in the process so that when I go to look at it, I can very quickly find out what that whole conversation was about.

(42:05) "

> > Cool.

> 

You should share that.

> > I'll I'll I asked

> 

> > to get connected to you on LinkedIn.

> 

I'll I'll I'll share that with you there.

> > Awesome.

> 

I'm in the the Slack group as well.

You can always find me there.

> > Okay.

> 

> > Yeah.

> 

And drop stuff, too.

Yeah.

With with Crewar, right?

Um my my thing was I wanted to make it as automatic as possible.

And so, one of the things I I had it do was anytime I land on a a custom GPT or use a custom GPT, it autoscrapes that that link and adds that to my library.

So, my library automatically grows as I continue to use them.

so that when I need to find that again I can it's already there.

(43:07) >> It's you know one of the things I found and you know you talked about

um using voice at least for me this is a truth in that if I don't capture why I'm doing something in the moment and then I go back to it a couple of days later or even a couple of hours later I may not

exactly remember.

So I have trained myself to immediately record and organize information in real time.

So later when I access it, I remember what it was for or even if I go is always a a source of discovery.

I I find that if you don't if you don't go through that process then

uh you will lose something in the process.

Yeah.

(43:40) Yeah.

Sounds like some of those principles.

Have you read the book building a second brain by what's his name?

> > Thiago 410.

> 

Yeah.

I mean that's what I've been working on that stuff for

> > 20 years.

> 

So that's uh Yeah.

> > Yeah.

> 

That's awesome.

> > I I resonate with what he says.

> 

### Audience question: Replit delivery, email reports, and integrations (44:04)

(44:04) Other questions?

> > I was wondering and thank you for your presentation.

> 

when you vibe code on replet for example like the example where you had it email over a report how do you pull that all together where you're vi coding the concept but then also wanted to execute the actual um whatever function like the delivery of it.

> > Yeah.

> 

> > Yeah.

> 

Uh honestly replet does most of the the heavy lifting for me.

So I tell it what I want to do and then it'll say here's some options and it'll be like connect this app which is an integration we already have inside a replet.

it'll tell me what it's what integrations are and it's like so go create an account and then log well log in here or give me your API keys and then it takes care of it from there or it it also depends on the client like if they're using high level as their CRM or whatever most of those have a way for you to push information into it via like a web hook and so we just have to set up a capture

(45:01) mechanism inside of the CRM and in order to capture that information and once we have that we just give that to Replet.

Replet's like, "Great, let me do my thing." And it does whatever it does to make sure that information goes.

And then it'll

The cool thing is it also will send test data for me um automatically.

So I can really work things out.

Um I don't have to do actual triggers.

I don't have to complete that quiz for example 20 times.

I can just be like, "Hey blah blah blah, hey replet, like send mock data now." And it will.

(45:18) And then I can test it out on on the workflow side of go high level.

### Crewar stacks: one-click open vs automation (45:39)

(45:39) >> I see.

Thanks a lot.

One other question about Krugar.

When you create a stack, what exactly happens like if you like will it automatically flow the content through or do you still have to go like step by step in your stack?

> > I would love for it to do that.

> 

Um, no.

Right now it's more of a one-click open.

It's kind of like when you need to open five tabs, like the group tabs inside of of Chrome, you can one click, it'll open all of them.

It's it's like that.

like I want to open all five of these custom GPTs at once and and it'll do that for me, right?

### “@ mention GPTs” tip in a shared conversation (46:38)

(46:38) And and go from there.

I'm still like like like I said, my process is kind of manual before I automate.

Usually, if I'm working with a team of of custom GPTs like that, I'll it's a lot of copy and paste from one into the next.

I play orchestrator, right?

And then that generates and does its task and then passes it on to the next.

Um, the reason I usually don't I I don't automate them until that copy and paste is perfect, meaning I don't have to correct what's happening with in that custom GPT.

I don't have to tell it to fix this or do that or whatever.

> > Hey, quick question.

> 

Do you know that you can share them?

You have them in the same conversation?

> > Yeah, I I do that a lot too where you do the at and bring them in.

> 

Yes.

Yeah.

> > Yeah.

> 

Okay.

So, limit the copy and pasting because I've ran into that issue as well.

> > Yeah.

> 

Yeah.

So, I'll do that.

Like for one client, like they wanted me to help out with their marketing even though that's not our focus anymore.

### Building a virtual “Director of Marketing” + specialists (47:00)

(47:00) Um, and I said, "Sure, I'll fill in the gap until you find your new director of marketing."

But, and that was the thing that we built a director of marketing for this functional medicine clinic.

That was the custom GBT, but then I'd bring in these doing what you're talking about, bringing in these other at um project managers.

I call them platform specialists whether it's analytics, solar, G4, Google Tag Manager, but all their experts on their different platforms and I bring them in to get their literal console and feedback in the same conversation.

### Audience question: finding new clients and pitching services (47:54)

(47:54) >> Yeah.

Yeah.

It's awesome.

Yeah, exactly.

No, that went away for a while.

It just came back.

> > Yeah.

> 

Yeah.

After 5.2.

> > Yeah.

> 

I was so sad when it went away because like this sucks.

This just made everything so much harder.

But yeah, that that's really simplified the process of creating content because you keep it, you know, once one stream or one conversation and I just bring them in and boot them out as needed, right?

> > So,

> 

> > perfect.

> 

Any other questions or

> > Yeah, I had a qu another question.

> 

Sorry.

> > Go for it.

> 

> > Yeah.

> 

> > How do you find new clients?

> 

Like how do you pitch this to new like as a service?

### Answer: referrals, discovery call, AI plan, partnerships, LinkedIn outreach (48:29)

(48:29) >> Yeah, great question.

Um we have been lucky in that we have a sister company called Wealth Factory and um so Exit Loop was a little side project inside of Wealthactory that me and a buddy created and then we spun off um with the owner's blessing who actually became one of our business partners.

And so we get loads of referrals from from them because they're coaches.

They each have about 200 clients each.

They're all business owners.

And so the coaches will then um recommend that they meet with us and we don't do a sales call with them.

We do an AI discovery and I just essentially sit down, chat with them, they tell me about their business and then in real time I'm I'm taking that that transcript or call into chat GBT and saying give me some ideas and then I give them three to five ideas.

(48:58) Um, and then I take that transcript and I have another custom GBT that actually takes the transcript and turns it into a 12-month AI implementation plan that's very high level, right?

It's just like a webinar.

You tell them what to do, not how to do it.

And so it tells them what they need to do each month and the estimated time savings.

And then I throw that into Gamma.

And then Gamma creates a beautiful presentation.

And then I send that to them.

And that usually converts.

### Company size focus and closing (52:01)

(52:01) I think uh out of a hundred meetings we have, we have 20 that convert into sales opportunities.

And then from that half of them aren't a good fit because they're too small, right? They're sub a million or something like that.

Um and so that's that's what we're doing right now uh to get clients.

But then we've also started be looking into partnerships.

And so partnering with um VC firms or things like that, people who are interested in fixing the operations of the company to increase its multiple so that when they go to exit, they can actually sell it for more.

And so we're building partnerships with those types of people with functional medicine influencers because that's that's a niche that we're we're growing in a lot.

Um things like that because they're they're tight-knit communities.

So and now we're testing just LinkedIn outreach.

Um, we're working with a guy named Phil Palucha who runs a company called Billionaires and Boxers.

Great great guy, great company, but we're working some of his his campaigns to do outreach and and whatnot.

So, we'll see how that goes.

We just started it this month and and I'll let you know for results and appointments or not, but but yeah.

### Wrap-up, links sharing, and end of session (52:39)

(52:39) Well, 

> > thank you, Pete.

> 

> > Yep.

> 

Hey, 

> > Peter, quick question.

> 

uh just 

> > is there a way to access the links that you were providing before and showing off?

> 

> > Yeah.

> 

Yeah.

I'll share this whole gamma slide with you guys and I've got all the links in there and and everything's um available and free like you can access it all, add it to your own accounts. You're you're good to go.

(50:33) This is way advanced from my knowledge of chat.

I thought I'd learn about custom chat GBTs, but like just hearing about how the use cases are pretty amazing.

Um, and the way that you're able to like screen do do interviews, you know, across many people and all the other use cases.

I have something a little bit maybe simpler, but um, building an interactive mobile gaming app.

Um, and like Sachbt or Gemini has been like guiding me along in terms of like what to charge and stuff like that.

Um, I guess that's something I'll connect with you.

I guess not.

I guess I'll connect with you afterwards.

> > Yeah.

> 

Um because I'd love to see like it did say, you know, for the first phase, you're better off using you're better off building part of it yourself.

> > Um but what size companies do you work with?

> 

> > Yeah, we work with 1 to 10 million.

> 

That's our sweet spot.

Those are usually and they're usually um local businesses, right?

So service industries, clinics, things like that.

And they generally are about 3 to 5 million in size. that's that's our our our sweet spot that we like to work with.

(52:01) They have enough bit employees to have enough problems for us to solve is what it is, right?

Um and cash flow isn't as big of an issue for them, too.

So, when we hit them with a 10K audit, they're like, "Okay, we're a 5K a month retainer," they're like, "We see the value, right?" Versus somebody who sub a million is like, "Oh, that's my child's college fund.

" You know, things like that. So,

> > right.

> 

Okay.

Very cool.

Thank you so much.

> > Yeah.

> 

We're out of time.

Uh, thanks everyone for coming, Pete.

This has been great.

It's so cool getting in the weeds and 

> > and and getting to see some of the kinds of things we hear a lot of people talk about but don't always get this kind of cander.

> 

So, uh, so I learned a ton.

I'm happy to share any of the links and references with everyone who not just 

> > uh attended live today, but folks in the community.

> 

I'm glad you're part of the community as well and 

> > folks know where to pester you for some followup tips and things like that.

> 

Um but this is awesome.

### Closing remarks and next week teaser (52:58)

(52:58) Like this is big big reason you know why we do all this just to be able to geek out with each other and and learn how to do a lot of things better.

So uh thank you.

Thanks everyone as always for great questions.

Uh got some fun guests. We'll talk more about the out of home space next week with Barry Fry from DPIA and 

> > and we've got a lot of great guests coming.

> 

So, thanks everyone.

Happy New Year and 

> > excited to see you way more this year.

>