How Creators Can Use AI Voice Agents To Grow Their Business In 2026

The next like 10, 20, 30 years, I don't think we'll be typing at all anymore.

Hi, I am Nathan's creator flywheel agent. What initially drew you to the Flywheel course?

There's a lot happening in AI right now, but something most people that I talk to aren't using enough is voice.

We're going to be able to actually get what we want out of the technology much more effectively.

My guest today, Jay Singh, is on the cutting edge of helping companies implement AI. And in this episode, we get really practical. You'll see how we actually interact with AI, the tools we use, and how we think about crafting prompts.

I think people should spend more time on prompts than they are.

How do you go from this like brain dump with voice into the desired output that you're ready to share?

When you're dictating, just also—

Yeah, I love that. Then we do something we've never done before on the podcast, a live demo. You'll get to see Jay show you exactly how he's building a working voice agent from scratch. And we explain each step along the way.

My workflow to create a prompt is—

I think that's fantastic. People should dive in, implement this right away. I love it. Jay, welcome back to the show for the third time.

Three-peat guest?

Yes, the one and only.

One and only.

So we're going to do a few different things this time. We're checking back in, but we also— we have our laptops out. You're going to do a little demo. Some live AI demoing happening. But first, when you were on the show originally back in the spring, we talked about the agency, how things are going, and a big goal. What was the goal that you had set when we were up on the board diagramming the rest of the business?

It's pretty crazy. It's only been 6 to 7 months since, since being back here. And yeah, we had the two, the two episodes. One was around AI and then the other one was you coaching me on how to be able to hit some of our revenue milestones that I've been trying to work on for the agency, our AI implementation agency. And the whole, the whole title of the podcast was called How to Go from Half a Million to a $2 Million Agency Business. We talked about all these different ways to be able to be more focused on your ICP and take a bunch of other advice that you've been being able to teach like a lot of other creators on how to build their own business. And since then, like we've actually gotten up to a $2 million run rate, uh, which is pretty cool.

We had that plan. What did we say?

I was like, yeah, maybe like a year and a half, 2 years. Yeah, exactly.

So it was by by the end of 2026, like, if we could pull this off, like, yeah, that'd be amazing.

So grateful for all your guidance. It's like very honestly, like, uh, totally changed the course of the business so far. And I think that's because of a lot of the thoughts that you gave to me and also just the tailwinds of what's going on in the AI space.

Of course, you know, we've been doing a ton of AI implementation internally and on our own. We've got a bunch of people who are very excited about that and like building custom apps or like one of my favorite examples is Sean, who's one of our creator growth managers. And, uh, he would have all these times where someone says, hey, you know, there's this Kit feature you'd probably never build because it's niche just to my use case, but we do this weird thing. And he's like, oh, okay. And then he goes and builds them a custom app on the Kit app store, right? Not to be published to anyone else, but he uses AI to do it on, you know, REPL or Cursor. And then he comes back like after the weekend and is like, hey, I built this app so that functionality exists. And they're just blown away because like, Who does that?

Yeah.

Um, so we had a lot of people implementing things, playing with it, but really wanted to bring a systematic approach to education and all of this. And really, I guess if it's some people are spiking really, really high, it's like, how do we raise the tide for everybody?

Yes.

And so you talk about the project that you came to do for our growth team specifically.

So it's been a ton of fun. Um, so the CRO of Kit, uh, his name is Dave. We've been able to work with him and the team to do a couple of things. So our main offerings are AI education, AI enablement, or AI engineering. And so what that means is education could be like trainings, workshops, like here's like the best-in-class prompts, here's like the best-in-class tools you should use. Enablement ends up being building light no-code automations on top of platforms like Relay or NN. And then there's also custom engineering work where it's like, you know, we're building out like an external-facing product or an internal product. Probably like, you know, more expensive, more resourcing, etc. For the growth team for Kit, we've actually been combining the first and second where again, working with Dave, his goal was, you know, we don't just want you to come and build automations internally. We want you to do that, but also be able to help teach our team how to be able to do that. Go, go look at what like amazing work that Sean has done. How do we, how are we able to kind of uplevel everybody else in the team to also start to identify opportunities to automate? Parts of their work? How do you even think about that? Uh, once you've identified it, how do you choose the right platform to be able to automate it on? How do you prioritize the right ones to be able to do? And then for the kit perspective, how do you then share it, uh, with other members of the team? So then they're also thinking about how to automate other parts of their work and kind of, you know, on the Nathan flywheel, like it actually starts like an internal flywheel, um, of identifying automations, building automations, sharing automations, and then inspiring other people to also do the same. Um, and so yeah, we've been, we've been working with the growth team, uh, to again, yeah, automate a few things, but also use those as examples in terms of how to be able to teach the team how to do it as well.

What's one of your favorite automations that, you know, has come out of this so far?

Yeah, so we, my favorite automations are ones that are modules that could be used for other automations. So that's a lot of buzzwords. Let me, let me try to break that down. Um, you know, when you're thinking about what to automate, uh, some of this stuff could be like very unique, discrete workflows that are just for that person, um, which is great. I think you start to get some, uh, interesting efficiency gains across the team when you can think about what are automations that if you, if you do that one thing, it'll help a bunch of other people do their job more effectively. And so the one that we've been kind of, uh, thinking through is like the, you know, content, the content creation agent. Obviously across the growth team, there's a ton of different avenues of, of content creation. There's creating case studies, there's creating like outreach material to creators. There's, you know, creating like other forms of like collateral. But there's one like foundational agent that would be really helpful to kind of create all these other workflows on top of it. So, you know, what are the inputs for that? How do we collect all those different data sources? How do we figure out the right prompting? How do we then, you know, coach the team to say, now that you have this block, you can go create all these other cool stuff based off of this input. Um, so that one's my favorite. Um, because it's not really one, it's actually just like a building block for a bunch of other ones.

Yeah, I like that. Okay. When there's a bunch of things that we wanted to get into, but when you were talking about coming back on the show, you know, we're like, hey, we're, we're hanging out in the same area. Like, let's, let's do another podcast. You know, what's most compelling, the thing that you wanted to talk about the most is voice. And I feel like there's a lot of fancy things going on in AI right now. And voice is one of those that doesn't get as much love as it should.

Right.

And so we're going to demo a bunch of stuff that you can do with voice and they're very specific. But why do you love voice so much in AI? And why is that like the underutilized tool right now?

Yeah. To kind of take a step back, I think if voice technology was created, let's say like 30, 40, 50 years ago, if we were just there from a technical perspective, I don't know if we'd ever even have to learn how to type.

If it was as good today as it is now.

I think we had to learn how to type because we had a constraint of a technology that was like, you know, you can't actually communicate with this piece of software or hardware that well through the spoken word. And I think if that was there, like, I think typing itself would have never necessarily needed to be created.

And our whole laptop that's in front of us would not have half of it.

Yeah. Like, what is that?

Right?

Like, and so I think, like, I think my, my take is that over the next, like, You know, it's hard to say timeline, but like the next like 10, 20, 30 years, like I, I don't think we'll be typing at all anymore. I think we'll just be speaking to our technologies. And I think that's going to be way more human-oriented. We're going to be able to actually get what we want out of the technology much more effectively. Um, and we're going to be going back to a piece of technology that we've been doing for a really long time, which is just speaking.

Right. Okay. So what's interesting to me is that AI has made it so that going from speech to text and text to speech is seamless.

Right.

And so you can actually interact with things in exactly the format that you want. So for example, I want to create content with voice because it's 2 to 3 times faster than typing.

Yep.

And I can do it in the car, on a walk, on my phone, you know, like all of these different formats.

Yeah.

I actually want to consume content.

Yeah.

Reading. Yeah. Because it's way faster.

It's faster.

Though, if like Apple Messages, I'm still waiting for their voice memos to be like 2x by default. So I don't mind a Loom so long as it's 1.5 or 1.75 speed or whatever. But what's interesting is you can go between these and we could each have our own preferences where before there was always some amount of loss that happened. As you went text to speech, it's like, eh.

Mm-hmm.

It's like 70% is good. Yeah. Now it's like 95% is good.

Right.

And speech to text is the same thing.

Yeah.

Like I was showing WhisperFlow to my 11-year-old yesterday and he goes, wait, what is this tool? And I was like, you know, I showed him and he's like, doesn't the iPhone already do that? And so I compared the speech in WhisperFlow and how it came out and how fast it was and how precise it was.

Yeah.

With what's built into you know, uh, iOS. And like we had the same, said the same thing twice, put in H1 and it was like, oh, WhisperFlow is just like near perfect text-to-speech. And so it's, or, uh, speech-to-text. And so it's interesting, like watching it, how does this change everything? So I mean, you have a lot of thoughts on that. So let's dive in.

I think, yeah, I think the, the Apple example is a good one to start with because I think when people think about voice, they just kind of write it off a little bit because their experience is Siri or an Amazon Home or Google Home or Alexa, and those experiences sucked. Like, it wasn't there yet because it was deterministic in terms of how that product would talk back to you because it needed to know exactly what you would say and have an answer to it. And it was deterministic. AI has— the technology is probabilistic. And so, you know, it doesn't know— you don't know what it's going to say back to you. But that also means that you don't have to like set these different, like, parameters of, like, if they say this, please say this. And so I think AI has unlocked the ability for it to actually feel more human, like, sound more human. I think that your, your, um, the, the story of, like, your, your kids being able to use it is a really fun one. I think, I think another element of voice is that it's, it's something that any age range could use, right? And I've been thinking a lot about, like, my parents. My parents are using, um, you know, they're, they're Indian immigrants. Like, they're, like, very well educated. Like, they type all the time, but, like, yeah, it's like, it's not as, not as crisp as, like, some of being able to just like be able to type everything. And so even my dad being able to speak into like advanced voice mode, uh, in English, but also his language, and then it kind of coming back to you in that same language, like that's really powerful. Um, so I think there's a whole generation of people that will be able to use technology like much faster, um, including those that are maybe, you know, like over the age of 50 that like, you know, typing may not be as natural to them. They can kind of loop back to voice as like that main interaction.

I think the other thing is that the, the speed of the interaction is interesting. So like WhisperFlow is probably my favorite tool on the Mac right now. And even if I'm working in a Google Doc or something else, I'm speaking probably 70% and I will type and speak and I'll be back and forth. But it's just, it's so fast and the interaction is so easy. Like actually I was watching another friend on a podcast demo a bunch of like development cursor and he was typing his prompts and I was like, oh my, you know, I was actually cringing cuz I'm like, no, no, no, download Whisperflow, press the, the function key and like just get, you know, these ideas out there so, so quickly. Um, so it's totally changed the way that I create.

I don't write anything that's important for me to write. I don't write it first. I speak to it first.

Okay, let's dig into that. So anything that's important, right? Cuz it used to be The opposite. It used to be the opposite. If it's important—

Or no speaking at all. Yeah.

Right. And I would type it out.

Type it out. Yeah.

Type it out. Like I'm thoughtful, I'm crafting my argument, all of that. All right. So talk about why is that? Why is speaking first more important?

So maybe work back from like, what does it mean for it to be important for me? Like, you know, a LinkedIn post, a memo, you know, to a client, content for a deck, you know, like a really important email. all of that starts with me just maybe not hitting advanced voice mode on, on ChatGPT, but even just transcribing it. Um, you could use, you could use WhisperFlow, you can, you can, whatever it is, but I'm just being able to communicate my free-flowing thoughts. And then, and then I can go and start to package it right afterwards. Um, and the things that I type for now first is things that I don't care as much about, like maybe like a quick, like Slack note to the team, or maybe like a text, like to one of my friends, like, you know, I'm not going to speak to that. I'm just going to text. Uh, but I think that, I think that is just fundamentally changing how we're using technology. And it, yeah, for some reason it's not being spoken about as much.

Yeah. I mean, the only time I don't use, um, speech-to-text is when I'm like sitting in the coffee shop next to people and I'm like, all right, I'll get out.

But now you can be on a call, right? It's like maybe it doesn't feel as weird anymore either.

Yeah. Um, so in that, a workflow that I do a lot is I pull up WhisperFlow on my phone and when I'm driving home from the studio or something, I'll just talk through an idea. Mm-hmm. But actually there's a place that my workflow breaks down and maybe you can help with Is I will get this brain dump and this perfect transcription of my idea and all of that, but it's stream of conscious. It doesn't, it's not formatted well. If you drop that into ChatGPT by default and say like, hey, turn this into something shareable with the team in a Slack message, it will. Well, actually, when I did this, because I was like, hey, here's two podcast episodes that I want to do. I'm riffing on it, you know, and ChatGPT is like, it was like bullet points and formatting and all of this stuff. It didn't have emojis because I already corrected that, like, at the, you know, larger ChatGPT settings level, but the prompt wasn't there. So how, how do you go from this, like, brain dump with voice into, like, the desired output that you're ready to share?

Yeah, I think there's different levels of complexity towards it. I'll give you, like, 3. One is super easy, then one is maybe a little bit more complex, and the other one's, like, a little bit harder to set up. The most simplest, the simplest one is just to, when you're dictating, just also tell it how to create that content as you're speaking. So at the end of it, I'm, you know, I'll, after I've just brain dumped, I'm saying, you know, I want to, I want to be able to also share this with the team, like make sure that my, make sure that the writing that actually comes out of this is like tight, it's coherent, it's clear. Uh, some sentences are shorter, some sentences are longer, you know, it's, it's simple and it's copy. Um, so please like, you know, structure it. And so you kind of teach it how to structure it while you're talking to it. So it's all in just one dump. And so you don't have to do it. Like ideally you don't have to do much more prompting on top of that. You're speaking the prompt. Okay. So that's like that, that's like an easy win. I think like, make sure you, make sure you, after you dump the idea, like remind yourself like where the output is and just kind of teach it how you want that to be created. And then I guess like a little bit more complex is if you just want to do the transcription, you can have like a preexisting set of prompts. I think people should spend more time on prompts than they are. Um, I think you can get a lot of effectiveness out of that. And then the third thing is, you know, you can share maybe previous examples of times you've done this from transcript to output of an email, maybe like 3, and then you can kind of preload that into like a, you know, like a GPT or even just copy paste all of it into like ChatGPT and then say, hey, now I'm giving you this input again. Here's a transcript. Here's how you did it before. Please do that again. Yeah, exactly. So that, that, those are like kind of steps in terms of level of complexity. The first one could work for the most part, you know, just talk the prompt.

So yeah, that makes sense. I hadn't thought about like putting the prompt directly.

Right. It's a weird, it's a weird shift, right?

You're like, you're just, you've been, you've been talking to you, but now I wasn't talking to you. Now I'm giving instructions, you know?

Yeah. Yeah.

Is there anything to make it pick up on the difference between that or it just does it?

It kind of just does it. So I, I don't know why I do it, but out of habit, I just make sure I put some dashed lines.

Right. I do that too, where I'm like, here's the section that I'm talking about.

I don't know. I don't know if it's helping.

Yeah. The computer's probably like, That's—

I got it. Don't worry about it.

Stop being condescending to me.

Yeah, I get dash lines. Thanks.

Um, so I want to dive in a little bit more on the content creation perspective, but then we're also going to demo like how you can make voice agents and, and so much more. Is there anything else on content creation that, you know, as we're going speech to text or any of these things, like for a creator, any other workflows that they should be using to make content much easier, much more easily?

Yeah, I think, I think you know, it's speaking is not necessarily just for text. It could be for other forms of output as well. Like the, these technologies are getting pretty good where as long as you have like a lot of context, which ideally is easier through speaking, you can then put it into other kind of products. So that could turn into writing output that can turn into video output. Um, you know, some of the Google models that are coming out are pretty crazy in terms of how good it's getting. Um, you could use it into image, you know, you can, you can, you can create a podcast from it. You can do all these other things with that. Initial set of, um, like long, like transcript from speaking, kind of like I was just talking about having, having like a foundational module, uh, for the, for the, um, kind of content creation work that we're doing for Kit. That's, this is kind of also treated for me at least it's treated as like a foundational module for anything else I want to do afterwards. Having that transcript of me just talking at length, whether me talking to it directly, whether having it coming from a transcript of a meeting, or for any of the creators having it coming from the transcript of like a YouTube, like a podcast, like any of that, it just is so valuable to create all these other types of workflows on top of it. So I think, I think that you can get pretty complex in terms of what happens after. I think the biggest thing is just make sure that at least upfront you have like those transcripts, put it in a Google Doc, just have it all listed out so you can do a lot of really great stuff afterwards with it.

Yeah. One example that I use a lot is over the years we've done hundreds of interviews with kit creators about their story, their background, all of that. And the hardest thing in writing, like writing a book or essays or anything, is like having the right example at the right time. And so Isa on our team, who's done tons of these interviews and, and told all these stories, she put all of them together in a single Google Doc. And then I uploaded that to NotebookLM because I think— does it still have the biggest context window? As far as searching or others?

Google generally has the largest context windows for all their models. So Notebook likely still has that. Yeah.

So basically it was the one that I found that was the best at ingesting like hundreds of pages of content. And then when I'm writing my book, I just say, all right, you know, pull in the section that I'm writing and then be like, find 5 creators from our source material that could be an example here. And it'd be like, oh, well, Terry Rice and Dori Clark. And it'll just like rattle it off. And it's not like going and grabbing some random thing. It's like from the 200 stories that you gave me, here are the details. And I'm like, oh no, that one doesn't quite fit right. You know, I'm running through these and I'm like, oh, these are great. And I'll pencil in a few of them. And it's taken one of the most complicated parts of writing and made it really, really easy because that's a unique advantage that we have is Kit has told hundreds of stories over the years. And so it's like, great, showcase those.

Yeah. I mean, one of my favorite books, uh, that, you know, went from a podcast to a book is like, you know, Tim Ferriss, like Tools of Titans. Right. Before AI, like, poor guy, my goodness, like the amount of time he had to probably go through and like find that specific quote that would latch, like, in a match to that specific idea that he wanted to share. But I don't know. I don't— yeah, it's like pre-AI, that would have been— that would have taken so much time and a big team and like to— hey, I have this idea, researcher, like, go find out where this quote was. Where—

yeah, now that's just, you know, now you just say, what are the best tools ever shared on this podcast in this format? And that, you know, exactly. What common threads, right? I mean, it just goes to show it's, it's easier to create content than ever before and it's more valuable than ever before, which you would think would be different. Like, right, I would expect that the easier something becomes, the less valuable it becomes because then it's flooded. It's, you know, the market is saturated or whatever. But the weird thing about technology right now is content creation is easier and that you get this compound return of, you know, audience, the insights, you know, your AIs are better trained. Like, it's wild.

Well, I think the asterisk there is if it's good content.

Yes.

I think, you know, it's easier to create content, um, and it's easier to be able to have that content be seen by more people if it's good content.

Yeah.

I think it's just, you know, you have really good content. And so ideally it does kind of rise, but I'm sure a lot of people are finding it really easy to create content, but that content is not that good.

Right.

Another piece on voice, like that is for me, like my most authentic story and narrative is when I speak first and then I write versus just typing because I'm riffing, I'm talking, I'm thinking through it live. For me at least, I, I, I find my, my, my like spoken word is like how I think through things better and, you know, how I communicate most effectively. And so the best content today that's standing out is one that are ones that is authentic. I feel the most authentic when I'm speaking. So if I'm speaking my written content, then, you know, it could share a better narrative. It can be more personal. Um, so there's another element to it of like, how do you, how do you, how do you stand out with written content? You just make sure that it's more authentic. And how do you create more authentic content? Maybe just actually talking about it might be easier.

Yeah. I love that. So diving into voice more specifically, you did something different when onboarding the kit team on this project, and that's that you didn't have a Google form. There were no surveys. Typeform, any of that. What did you do instead?

Yeah, so the way we like to start any engagement is by running through probably like a 2 to 3 week discovery sprint. We go talk to the leaders on the team. We understand, you know, what are the pain points that they currently see in the company? Um, what are the kind of key, uh, priorities for their business for the year? And then also like what parts of their workflow, um, are, are so much of a pain that it could, you know, AI could be able to help automate some parts of it. Um, so we did that with the leadership team. but it would have just taken some time to be able to talk to everybody else on the team. And if we did, you know, it would have been like a quick 10, 15 minute chat and, you know, it wouldn't be, it wouldn't have, it wouldn't have been as effective. And so what we did with the kit team is we, we did the leadership conversations, but we also created a voice agent on ElevenLabs that, uh, was prompted to ask similar questions that we would in a discovery session with the leadership team, uh, to be able to send to the rest of the team. And so we sent it out to, I think, 25-some-odd people on the growth team. And I think we had 22 people that that submitted responses to it. And the learnings there were pretty cool. We saw that the average conversation length to the voice agent was 10 minutes. And we also saw that a couple people even spoke to it as long as 45 minutes. And what we heard on the other side of it was an incredible amount of context and insight that immediately gives us ideas in terms of what to be able to automate for you all. Because they were saying things like, you know, whenever I have to, you know, use these different systems and have to create this thing, like, I wish I was able to, you know, it was very personal, it was very authentic, but then we can look at that information and say, okay, cool, like, you know, these people are saying this thing, here's the common themes, like, hey Dave and the growth leadership team, I think you should automate these pieces. And so it just accelerated our ability to get to that kind of recommendation faster because people, uh, were speaking to the voice agent versus, um, maybe needing us to like have scheduled time to go talk to everybody.

I'm realizing as I'm working on my book right now, working on Ladders of Wealth, I want way more stories from people. And so I should be making a voice agent that is open. Like, I should send out an email to my audience and be like, hey, will you do an interview with my voice agent? Yeah. You know, I'm like, I'm doing a bunch of interviews with individual people.

Sure.

Right. But I can't do that at scale.

Exactly.

And so, right, you know, I sent out this email like, hey, what were the most important skills that you've learned in your career? Where are you at in these ladders? What helped you make the jump?

That'd be so cool.

If you were giving advice to your 18-year-old self, like starting your career over, what would you say?

So cool.

Um, you know, where, where does this framework break down that doesn't make sense to you?

Yeah.

Like I could collect hundreds or thousands of responses quite quickly.

Yeah. That's, that's like what I think people, when you think about AI broadly, you're like, it's going to replace, you know, my job is going to replace like what I do. And there's, there's a risk to that, but it also creates totally new opportunities that you just would not have had the ability to do. Like, how would you have, how would you have gotten this information otherwise?

You'd hire researchers.

You had, you had a lot. And even, even that, the amount of scale that you'll be able to get from researchers versus scheduling and everything else. Scheduling. Yeah. It's probably like, even if you hired a big team of researchers, it's probably still like an order of magnitude different in terms of how much information you can collect from people. And they're engaging and they're, then they consent to it and they want to be able to share. They want to be understood. They want to feel heard. Which is, you know, like it's this whole other element of it too.

So there's a bunch more in voice that I think is super interesting, but we've got our laptops out.

Yeah.

So why don't we just dive in and why don't you show us how to go and create one of these agents?

Let's do it. Let's do it. So what agent would you like to create?

Let's make one that would be relevant, I think, to any creator who has a course or community.

Okay.

So you're— you've gone through my Creator Flywheels program. We've implemented a lot of stuff from there. I'm launching the new version of it soon, and I want to collect feedback from all of the existing members on whether they love, what's missing, all of those things, right? So I can launch in the next version. I launch is the best possible.

Amazing. So maybe just take a step back. The voice agents today are actually pretty simple to grade. You know, you don't need a developer. You don't need, you don't need, you really don't need much technical skills at all. You can use ElevenLabs. There are other platforms that are, that are, that are doing this, but ElevenLabs for me is like the simplest. So here in this interface that you're seeing, I just went onto the ElevenLabs platform, signed up, and then, you know, I just came and created an agent as well. So now I'm trying to create like a voice agent.

So, okay. So you selected business agent.

I actually just created a blank agent now. Yeah. But you can use, yeah, you can use, you can use like a template that they have too.

Okay. So blank agent, you named it.

I named it Nathan's Creator Flywheel Agent. Um, you know, we could work on that copy a little bit better, but you know, uh, you could, you could create it so that it's voice, uh, or it's chat. So it doesn't have to be voice, but we want voice. We want voice. And so, you know, creates an agent. And so there's a bunch of stuff going on here. Uh, honestly, you know, The most important thing to do is frankly just figuring out what the first message should be and then also what the system prompt should be. Would it be helpful to kind of walk through these two? This could be simple. Maybe what would you like the agent to start off the conversation to your flywheel?

So what I want is, this is a targeted agent, right? So we're talking about We're collecting feedback from existing students. So maybe thanking them, like, thanks so much for investing some time in the feedback. Uh, or like, um, you know, taking some time to help me improve the Creator Flywheels program. And then just ask that first question.

Perfect. So, you know, why don't we do a little meta piece here? I'm actually going to use, uh, one of your favorite products and one of my favorite products, Whisperflow, to be able to just speak into this here quickly. So Why don't you like, okay, so I'm just going to try this right now. So, uh, your first message should be, uh, to be able to capture the person's name, uh, to be able to thank them for being a part of the Creator Flywheels course. Um, and to, uh, just kind of like let them know that they're going to be kind of going through this experience to be able to give feedback on how to be able to improve the experience for them.

Cool.

So this is, yeah.

Um, and so here you're not giving it a script, you're giving it really a prompt.

Yeah. So this one wouldn't probably go to the first message. It'd probably go here to the system prompt.

Yeah, exactly.

Good, good call. And then this is like quite literally what it'll say, right? So what is your name? Yeah. Something like that. And so, oh, actually here, here. So you intro, this is like you introducing yourself. So you can say, one sec. Um, yeah. Hi, I am, uh, Nathan's, uh, creator flywheel agent. Um, I'm gonna be able to better unders— I wanna be able to better understand, uh, some feedback to be able to make this experience better. Uh, first, can I get your name? Kinda, it's, it's, you know, you have to flip back and forth between like, am I the AI?

Am I?

And so, right. Yeah. So this is gonna be the first thing that it gets. On the other side of the screen. Do you want to make any changes here?

No, that sounds good.

All right, cool. So now I'm going to create a prompt. So my workflow to create a prompt is, let's see, maybe actually we can say some templates here. Okay. So here, you know, there's a couple of kind of key elements of a prompt. You want to give it some context. I'm just going to type this up right now. give it, um, a style, uh, guardrails, uh, types of questions to ask.

Um, and actually one quick meta thing.

Yeah.

I would type, I would speak.

Yeah, exactly right now. Each of those things. Yeah. And just that, yeah.

Part of Whisperflow making it so easy to, is to jump in and outta transcript mode.

Exactly.

That probably if I'm typing more than 2 words, and just go Whisper Flow. Then I just press the function key and trigger Whisper Flow.

Yeah, exactly. You're more of an adopter of this stuff than I am. I love it. I love it. And so why don't we, why don't we just do this right here? So what I like to do here is just like kind of go to ChatGPT and, um, be able to help it create a prompt for me.

Okay.

Um, so here's, so here's what I'll do. I'm going to, I want to do the following. Um, I want to be able to create a system prompt for a voice agent that helps to be able to speak with existing, uh, participants of Nathan's Creator Flywheel course to be able to better understand how it can improve the experience of the course, uh, for kind of future experiences. Um, uh, walk through, please like distill out a prompt that I can then share to my voice agent, uh, specifically, uh, for the context you have, you understand the context, uh, for the style of the agent, make it friendly. Ask follow-up questions. You know, don't be overly nice, but be specific and detailed and just like continue to be curious. In terms of questions, what kind of questions would you want to ask?

I'd want to ask the biggest pain points that you had in implementing Flywheels. If you had an aha moment, if you're giving advice to someone else going through the course for the first time, What would you say? Let's start there.

So I hope you captured that was for Flow. Um, let's see if it did or not. Yep, it did.

So perfect.

That's, it's prompts all the way down. So I'm trying to get, I'm trying, I'm trying to prompt ChatGPT to figure out what to prompt the voice agent.

Yeah. So it's splitting out role and tone and it's saying, yeah, you're like, don't be overly nice. 'Cause that is a thing where it's like, "Wow, thank you so much." You know? "Thank you so much." "You're the smartest person I've ever talked to." Relax.

Relax. Um, any guardrails you wanted to put in. So these are things that you wanna make sure the agent does not do. Outside of being overly nice.

What types of guardrails would you typically put in?

Yeah, that's a good question. So, um, I think, like, depending on the context of the agent, it could be things like, try to wrap up the call in a certain amount of time, right? Because sometimes it can just go on for like, depending on, maybe that's a good thing, right? I can actually share some guardrails after as well. But yeah, the level of complexity increases as you kind of figure out the right use case. So if we're building a voice agent for a Fortune 100 brand and we're about to put it on their website to interact with customers, Right. A lot of guardrails in place there, right? Um, you know, make sure you do not like, uh, offer like these amounts of discounts. Like make sure that if they say this, then it connects to a person. Like, um, you know, please look at this other file that has all of our brand guidelines and just like, you know, confirm that you can align with that. So for this, it feels like, you know, um, if you wanted to have a more sophisticated perspective here, maybe, maybe you have brand guidelines or something for, for Kit or for the Flywheel course you can put into here, but For this use case, it feels like it's okay because it's more of an interviewing space.

We're collecting information.

Yeah, exactly. Um, so again, when you get up to like these, like, like the bigger companies who have like a lot more kind of like legal requirements, this is where probably you'd put those guardrails.

Okay. So for now, so a guardrail might be focus on asking questions and learning rather than giving advice.

Uh, please, please create a guardrail section here. And one of the guardrails is focus on asking specific questions versus giving advice. Anything else while we're here? Great. Awesome.

And it comes back with the whole thing with the new guardrails section.

Mm-hmm.

Stay curious. Keep responses concise. Don't overphrase. Yep.

Yep.

Yeah, that all looks good.

Okay, cool. Sweet. Let's just see how it goes. So I'm taking this prompt, I'm coming back here. Um, it's not formatted that pretty, but I think it'll—

I don't think the AI is going to care.

Yeah, I think it's just humans who care about that. Um, and then, you know, I like to just make sure when I'm— I can choose like, um, you know, I'm— if it says fast and it's not that I'm going to choose fast for now. We can optimize on all the different models here, all the different things.

What do you look for in choosing the model? Should a layperson care about one model over another?

I think for voice, the faster the better. And realistically, this one is more expensive than others, but if you look at how much more expensive it is, unless this is being sent out to millions of people, it's not going to matter. And even if it's being sent out to millions of people, like, you know, I can, I can help you estimate how much that will be before you roll it out, but it's still going to be pretty, pretty cheap. Um, and yeah, so I, faster the better, because if it's slow, that's when it feels like it's a voice agent versus just like a conversation, you know? Uh, temperature is just like, you know, uh, you know, we'll kind of allow it to be a little bit more, you know, not, not super random, but like maybe give it a little bit of personality. And then, yeah, make sure you can kind of end the call. I wouldn't worry about anything else here. Okay. And then, and then what I would do here is just like start testing it. But instead of testing it, I'm just going to— we can just interact with it right away and see how it is.

So yeah, we could start collecting information because you've gone through it. The Creator Flywheels course.

I'll be, I'll be on the other side of it. Let's try it out.

We'll be the guinea pig.

So, I mean, that was it. Like in terms of, in terms of the creation flow, right?

We spent more time talking about individual form fields than like it actually takes to make it.

Yes. And so that, so now we'll interact with it and likely, you know, as we start, there's going to be some questions that it asks that are not the best or, you know, now that now we go through the testing process, right? And I think, I think what's changing in terms of building out products in the AI space is that it still may take the same amount of time to get from 0 to 1, but the actual creation of it is much less time, but the testing and the iteration on it should take a lot of time, actually. But the time itself—

we might be 20 minutes to get it live and then you might spend hours.

Yeah.

Like really refining.

It took 3 minutes to go live. Right. But, you know, it may take another like, yeah, hour to kind of like refine, you know. So let's just try it out.

Hi, I am Nathan's creator flywheel agent. I'm going to be able to better understand some feedback to be able to make this experience better. First, can I get your name?

My name is Jay.

Thanks, Jay. What initially drew you to the Flywheel course?

Probably just Nathan Barry. I mean, he's a pretty cool guy. He's building a pretty big business. I love how he talks a lot about creating content, but also being an entrepreneur. And as someone who's trying to kind of embody like a similar lifestyle, I thought it'd be great to be able to learn directly from him.

It's great you're inspired by Nathan's entrepreneurial journey and his focus on content creation. What specifically about his approach to those topics resonated with you and made you want to learn from him directly?

Yeah, I think, I think the way that he thinks about flywheels is really interesting. It, you know, while, while that concept, I've heard about the concept before, I think the way that he's putting a lot more precision around it has been helpful. Um, showing a lot of case studies, showing a lot of examples, and then even in the workshops itself, kind of, you know, helping, helping me figure out how to also apply it to my own business. That was really valuable upfront with the content that he was creating, but also in the classes itself, I found that really valuable.

That's really helpful to hear how you found value in both the case studies and the practical application within the workshops. Thinking about the implementation process, what was the biggest challenge you faced when trying to apply the flywheel concept to your own business?

So maybe let's pause there because I noticed a few things in the agent that I would want to tweak.

Yep.

Which is, I think, what we're going for. Yeah. I would want the reflection back to be shorter.

Yep.

Right. Because, you know, more of a natural conversation. Yeah. So you might say like only reflect back half the time or try to make that as tight as possible. You know, being a good listener, that's part of it. But also it's like seeing the formula to it.

I know, I know, I know. We've experienced that before too. It's like sometimes you're like, please just like stop like being like a goody two shoes. Right. And sometimes it's like that, that's where like a lot of, that's where like sometimes you have to kind of spend more time refining it. But yeah, like we've experienced, we've like, please do not ask all these follow-up questions. And sometimes it still does because it's, I don't know, I guess it's the follow-up questions you want.

It's the reflection, like the active listening. You're like, please be a worse listener.

Please be a worse listener. Yeah, exactly.

Or like, I know you, you understood it.

Yes, exactly. Exactly.

Yeah.

So, but those are small improvements that we can keep making.

Yeah, exactly. And then, you know, you could tweak the opening prompt. You could do a bunch of those things.

Totally.

So then in this, um, agent, what's happening with, like, where do we go from here? We have the agent and then what are we doing with the information that it's generated?

Totally. So, so here, like, you know, I, it should, I should have kind of recorded this, but I'll maybe like refresh it. And so once you're here, you can kind of come into evaluate. And so look at the conversations. You see Nathan's Creative Flywheel agent. We have a bunch of different agents. And yeah, you can see the transcription right here. And then you can copy the transcript and then you can kind of go work with it wherever you want to afterwards. So I would do something like this. Obviously it's not going to be super detailed, but I would say something like, hey, I just had one person be able to fill in my voice agent. Do you mind kind of in really crisp detail summarizing the key themes and pieces of pieces of feedback that were captured to improve my flywheel course. You know, something like that. And then I would just dump the transcript and let it kind of run its analysis.

Yep. And in this case, it has only one example, but it's already starting to pick out some of those themes. Okay. So I don't want to do this manually. I don't want to pull the transcripts back and forth. I want the transcripts to automatically go into, I don't know, like GPT or something like that. Right. What would be your go-to tool for making that happen?

So you could do elements of it here. Um, it's, you know, they're, they're continuously like kind of like building out their product, but I would come to integrations here and then you can set up your own like custom MCP server, which is, you know, it can get a little bit complicated, but the TL;DR is like, you know, you can come and connect this to wherever you want to send it. Um, some of this stuff might you know, kind of take a little bit of code. Um, but over time, like, they'll likely, they'll likely be able to set up a simpler integration here. Um, similar to kind of how on ChatGPT you have like, you know, add files or add from Google or use connectors or whatever. Like, you know, I think, I think over time, like this, this one is showing right now straight into the project knowledge.

Exactly.

Yeah. So for us, we, we like to keep everything in like a Google Drive. Like I think just having everything in a Google Doc is all of these.

Integrate with Google.

And then I can use those for future other workflows. Like we were talking about earlier with like having like that content module, like for us, we need to have everything that's in like a Google Doc. And so, yeah, like for this specific thing, like, you know, they, they, I think they have on the roadmap, they're going to have a little bit simpler integrations, but here, here's where like, you know, if you need help with this, like we can, our team can kind of just like set up like a custom MCP server to send the information wherever you want it.

Does 11Labs integrate with Zapier or Relay or one of those? Um, or would you have to do it through yourself?

You have to do it. Yeah. Right now you'd have to do it custom. Uh, it's actually, it's, I mean, and you're a developer, so I know, you know, it's like not overly complicated, but you know, this stuff might be at least right now a little bit complicated, but this will shift over the next like quarter or two where it should just, it should just send it over there automatically.

Yeah. Yeah. It's the thing is, can AI do this? And you're like, not yet. And yet is measured in months.

Yeah. Single digit months, pretty much always. Exactly.

And so in this case, as we ran through it, right, you know, our use case is I'm relaunching the Flywheels program. I want to know what to do better. Yeah. Now, like, I could have this library of, say, I send it out to 40 people, maybe 20 people respond, have conversations with it. I could collect a bunch of information. Now I can write my marketing copy off of it. I can make a custom GPT that has my, you know, other prompts, my material, and then also now has the feedback.

Yeah. Yeah. I think, I think the flywheel to think about if I'm a creator, um, maybe do you want to, do you want to maybe like, um, share like a creator, uh, to kind of walk through how they'd be able to do this? Like maybe it's someone who's creating a course or who comes to mind and I can, I can kind of, you know, kind of give recommendation on how they could use this. Yeah.

So I think like if we use Jake Klaus, right, he's been on the podcast before. He has this amazing community called The Lab and, you know, he's, it's very, very active. Um, And so he might be thinking about what does he launch next? Mm-hmm. Right. I think that's a common thing, right? You know, what do I launch next or how should I level this up? Yep. Yeah. What comes to mind?

I think, I think for Jay, uh, you know, it's, yeah, coming in, doing a very simple kind of voice agent creation here and saying, hey, um, you know, here, uh, to the, to the community of how many people are in the community? Like, let's say it's 500.

I'm not sure exactly.

500 people in the community. It's like, hey, we're just collecting some feedback. Uh, we want to see how we can improve our services and our offerings and see what else I should be able to build. Here's, here's like a voice agent, uh, that you can go talk to. Um, if you think that's weird, here's like a Google Form, right? Right. Like, like when I sent, when I sent this out to the team and when you send it out to clients, uh, to like get information during the voice agent, there's like this weird transitionary period where some people will be like, that's weird. Yeah. Um, but then when you try it, you're like, that was actually way better than a Google Form. And so what I'd recommend tactically is like, just offer both, um, more as like a, homage to like, hey, if you need, if you need to go back to this thing, like, go look at it. But then when you've, when you do the voice agent and then you open up the Google Form, you're like, what? Like, how did I do this before? Like, when, how is this collecting any relevant information, right? You're like, 1, 2, 3, 4, like barely text any, type anything, you know? Like, it's, it's been a terrible experience.

And like, people, people make decisions off of that. I would probably send out the email. Yep. Like, let's say that I maybe I'm going to do this, I don't know, doing the research. How would I do this? So I'm thinking about the research agent for the Ladders of Wealth post, right? So this is maybe going out to a bunch more people. Sure. But I filter my list in some way. So it's the people who are more engaged in my content, you know, not someone who signed up yesterday. Right. That kind of thing. Yeah. And so maybe I send an email out to 5,000 people. Yeah. And the subject line is just walk and talk. Yes. I love it. Hey, could you do me a favor? I'm doing research for this book. If you meet these criteria. Yeah. Would you like next time you're out for a walk or all that, step away from the computer, take your phone. I love it. And just interact with this agent for 20 minutes. I love it. Like one, you'll get your steps in. Two, it'll really help me with this book that I'm working on. I love it. And, uh, you know, let me know how it goes. Hit reply and let me know how it goes. I appreciate your time.

I love that idea. And like this, this also kind of connects to like, you can, you can pretty simply, uh, connect this simply, you know, you can connect this to Twilio. Okay. I just set up a number. And so the voice agent has a number. Uh, and you just pick up the phone call. Yeah. And so, and so like we, like, you know, we've created an agent that goes and interviews people. Like we differ kit, but we also have a number that you can call too. I think it's like, you know, I can, I can share it later. It's like 1-855-4-CASPER. And like, you know, it just like you're talking to it just like you would just, uh, sign up for an 11Labs agent here. Um, and so it is the agent, but it's just like connected to your phone. And so for that, I would even recommend just call. Just call this number. And like, when you call, it's like, hey, like, this is Nathan Barry's, you know, agent. Like, uh, how's your walk going? Like, you know, let's go for a walk and like talk about, talk about the book and talk about stuff. Right. So that, that would be a really cool way to implement it.

I like that. Something else that stood out to me about voice and really these agents that are separate from the creator is how people interact with them differently. So at Craft and Commerce, the team from Delphi was out and they're building these clones where it's like, hey, it'll answer questions and like, Our examples here have been research and collecting information, you know, but you could obviously build stuff to give advice and, and do so much more. Right. Um, that's what Delphi is really focused on. One thing that stood out to me was Jason Lemkin, who runs the SaaStr community. He launched his Delphi clone and he had this like 10 things I learned since launching the clone as a post on X. And the one that stood out the most is he said that people would ask it questions that they wouldn't ask him. So they were asking him questions like, hey, how do I reduce churn by 2 points? How do I raise the next round?

Smartypants questions. Yeah, exactly.

And then when they talked to his agent, they would ask questions like, how do I break up with my co-founder? What do I do when I don't feel like I can keep going in this company? How do I know when to call it quits? These things that are like the existential problems that every founder deals with. Yeah. But if I ask Jason that live, he's like, oh, so the company's not going well, right? There's like this ego status thing going on. Like, how do I ask Jason a question that makes me seem smart? Totally. But then when that's totally gone, then let me just ask the question that like I'm actually really struggling with. Right, right. Yeah. How do you see that playing out further in, you know, all these voice and agent interactions?

Yeah. I mean, it's a, it's a, it's a very unique kind of odd, like, you know, realization that you feel sometimes more comfortable talking to a voice agent than another human, right? Like, or like a clone of that same human. Um, I think it, it's for a couple of reasons. Like these, these agents are trained and just AI, like these models are trained to be like quite friendly and kind and, um, and they ask like really good follow-up questions and, um, you know, and like the lack of judgment is definitely there. Uh, and so it's, it's, it is, it is just like a very nice person or a nice like experience to have. Where in reality, most people don't have those experiences with other human beings a lot. We're communicating with a lot of really great people and we know a lot of folks that care about relationships and asking questions and developing a bond with folks. But a lot of people are quite lonely. A lot of folks don't have a lot of people to go talk to. And so when they have these AI agents that are just talking to you, Again, it's a little bit odd because you're like, well, why don't you just go talk to another person? But that may not be an option for a lot of people. And this voice agent could just be actually asking the questions that no one's really asked you before. And so, yeah, again, there's an element to it which is a little bit concerning because you're like, are people just going to be talking to these voice agents all day long? And what's the negative impact for that? But yeah, the lack of judgment definitely helps. You know, if, if Jason was like on a back-to-back call and asking, like he was tired, he like, maybe he doesn't really care. Like he's at the end of his day or with his agents just like on all the time. Right. And so, um, where the participant really wanted Jason to be active, they got it with the agent.

Right. Okay. So let's talk about voice and where it's going with creators or what other things should creators be implementing as it relates to voice agents or, uh, other workflows?

Yeah, I think this one, this one is tactical, um, where it's like get feedback from your audience. Using a voice form, like instead of like a Google Form or like a, you know, like a traditional form. Um, that way you can kind of create like some level of a flywheel where you can collect like really, let's go back to Jay's example. Like you collect feedback from the community. Um, let's say you have 100 people who like submitted some responses that then can be created as content for you. Um, that then can also just like guide your decision-making in terms of what's the next service or product to be able to build. And then you can then repeat that process again. So you get better ideas, you get a better pulse in terms of what your audience really wants by sharing these voice agents for feedback collection. Not even as a clone, it doesn't have to be you. It can be a generic voice agent if you're feeling uncomfortable about that, or it could be you cloned yourself. And then the other piece is just for you to be able to expedite your own content creation flow, whether it's writing, video creation, podcasting, et cetera, that original idea generation, um, to, to have voice to be able to kind of help you unpack certain ideas, I think is like a, a really good way to start. So those are two very, um, simple tactical things to just do this week and like experiment with this week. Um, you could set up these voice agents quite simply and you can always just like dictate yourself to Whisperflow or to ChatGPT. I love that.

So what I'm hearing is two very practical takeaways. One is incorporate voice as much as possible in your content creation flow. 'cause you can make it faster. You can riff on ideas. You can probably get to higher quality concepts, and you can do it when you're doing the dishes or driving or, you know, any of, any of these other things. And make sure that you're storing that information that you're putting out in a place that you can refer back to, prompt off of all of that. And then the second one is to incorporate voice into your research process, because I think that most creators do probably a 10th of the research or a 100th of the research as they should. I know I do. Right? And because you just can't do it at scale in a personalized way really until now. And so by incorporating voice into the research process, then you could have so many conversations. You could pull up, you know, like source from dozens or hundreds of people and then really find all of the common threads and the best stories and everything from there. And you can take all of that to then bring it back into your content creation process. Probably using voice again in order to get to the finished product. Is that right? That's right. Okay. So a couple other things that I'm thinking about before we wrap up. One is that in flywheels, there's often a small tweak that then creates another byproduct or really closes the loop. So if we're thinking about like the content ideas flywheel, that's one where you're in your newsletter, you're asking questions and like, hey, what's your biggest struggle? And then using that to create content. You could incorporate that into your voice agent where you're just saying, hey, what's your biggest struggle related to learning flywheels, you know, learning how to run Facebook ads or that kind of thing. And that would get you like very specific things to make content about. Yeah. Or the other one is you collect testimonials. I'm thinking about for the Ladders of Wealth essay, I could say, you're like early on, I might ask, hey, have you read the essay or not? And it's like, yes. Then later we could have in the prompt, like, hey, if they have read the essay, wait till later on, say, hey, what stood out to you most in it? Or did that have an impact on your life? Like I got this unsolicited testimonial from a guy named Jay Yang who wrote the book You Can Just Do Things. And somehow I'd posted about the essay and he replied and he was like, reading that essay accelerated my career by 3 years. And like, now I'm interviewing Jay to be like, all right, what specifically? Like, You know, make sure to emphasize that. But if I— in books, you're really trying to get to these portable stories, like what's the thing that resonated and stuck with someone? And so that you can then package it precisely in a way that they'll remember. Right? That's what James Clear did with Atomic Habits, right? 1% better every day. Every action you take is a vote for the person you want to become. He's got a couple of these that people remember, and that's because he's obsessed over making them portable stories. And so if in this interview process I can get people to reiterate the things that impacted them and the things they remembered, then I can be like, all right, here are the 8 things that I need to figure out how to, like, get the messaging on absolutely perfect. Because that's going to be the ideas that spread and that's going to make the book do well. And I would spend hundreds of hours doing that if I did it as a human. Or I could spend, you know, a handful of hours doing it with AI. Yeah.

And a lot of people want to likely have some sort of conversation with you and just be like, Nathan, thank you. Right? Like, I've had the privilege to like do it in person a couple of times, but like I think a lot of people do. And that obviously will create really good testimonials, ideas, content for you to be able to kind of package up what that, you know, shareable story could look like.

Yeah, I love that. All right, Jay, so we've covered a lot of stuff. People should dive in, implement this right away. One, where should people go to follow you? And then I have a follow-up question after that. Yeah, LinkedIn. LinkedIn's the spot.

Jason Casper Studios, LinkedIn. Yeah, my email is jcasperstudios@xyz.

And then the main things that you help people with, like Kit's a Client, you know, uh, that AI implementation, what other thing, like what's the breakdown of the 3 main services?

Yeah, so we do AI education, AI, um, enablement, and then AI engineering. And so you can think of that in terms of, you know, workshops, trainings, uh, teaching the team on how to be able to use AI more effectively. or that could be actually using like no-code automations to be able to have like simple workflows to, to integrate into your business. That's kind of what we call the enablement. And then depending on how complex the organization is, we kind of go into like actual like custom engineering. Um, so yeah, we're working with, we're working with Kit, which has been awesome. Um, other growth stage startups like hedge funds, public market companies, and then obviously a lot of creators as well. Sounds good.

Well, congratulations on, uh, 4x in the company between when you came on the podcast before and now. And, uh, we'll have to have you back for the, for the fourth time at some point in the future. I love it.

Thank you. Thanks for coming on.

If you enjoyed this episode and you love getting a glimpse into Jay and the way his mind works, check out episode 76. That's where we get up onto the board and we break down how to grow his agency from $500K in annual revenue to over $2 million.

I thought it would take a year or more.

He pulled it off in under 6 months. And that just is a testament to the plan we came up with and just how good he is at executing. So check that out. Thanks so much for checking out this episode. Be sure to like the video, subscribe, and I'll see you next week.