The AI Coach
AI meets human behavior in these fun and insightful conversations with Danielle Gopen, a top founder coach and advisor, and Paul Fung, an experienced AI founder. We blend AI breakthroughs with real-world insights on business strategy, industry disruptions, leadership, and psychology while cutting through the hype to ask the hard questions.
This is for decision-makers like founders and business leaders who want to harness AI's potential thoughtfully, exploring both the technological edge and human elements that drive long-term success. Join us to challenge the status quo, navigate the shifting landscape of an AI-powered world, and get new information worth sharing.
The AI Coach
Using AI to Build and Launch A New Product
Have model anxiety where you don't know which LLM to use and why when a new one is released? Fear no more, Narrow.ai is here! Paul talks about his company's launch (out of stealth mode) and shares how much they were able to use AI to do their product launch, saving them tons of time and money. We also talk about non-technical founders using Claude as their CTO to build new apps. This is the future of using AI to do more faster.
Links and Resources:
Narrow.ai https://www.getnarrow.ai/
Soft Landing Meditations https://www.softlandingmeditations.com/
Hoop https://www.hoop.app/
We love feedback and questions! Please reach out:
LinkedIn
Episode Directory
Mentions of companies, products, or resources in this podcast or show notes are for informational purposes only and do not constitute endorsements; we assume no liability for any actions taken based on this content.
there is so much to talk about today, danielle. So much has happened. We've been gone for two weeks, little hiatus.
Speaker 2:The people have missed us and now we are back to bring them all the latest ai happenings and, as we can imagine, a lot has happened in those two weeks, both for AI overall, but also for you, paul.
Speaker 1:Yeah, very excited. So last week, jumping right into it, we launched a new product. We actually rebranded our company, launched a new website, redesigned our app and launched a new product. So super excited about that. That was really busy and I do want to get into that a little bit, but I think one of the reasons I want to get into it is because one of the things I was hoping we could talk about today that I thought you would like a lot, is how AI is changing entrepreneurship, just like how it changes startups.
Speaker 1:How quickly we can move because we're able to write code faster. We're able to, you know, get designs done faster with some of these image models. We're able to write marketing copy faster. That would have required marketing teams and turnaround times and stuff like that, and so we used a ton of this stuff for our own launch. I also have a friend who launched a product last week as well, and they have a very cool story. They basically used AI as like a CTO, and so they actually used it to do all the coding for their website, which was amazing. So I'd love to talk about that. I thought you'd find that interesting. And then, yeah, there's a bunch of other stuff that happened this week related to image models, and everyone on Twitter is going crazy about Flux, which I've been playing with today, and it's absolutely insanely good, mind-blowingly good. So, yeah, where do you want to start?
Speaker 2:Well, what you were just saying about the launch and all the ways that you used AI with it made me think of something I saw I don't know today or yesterday, when somebody wrote everybody's afraid of AI taking their jobs. What you should really be afraid of is losing your job to other people who understand how to use AI.
Speaker 1:I love that. Yeah, that's so good. That is so on point, because the people who know how to use AI effectively are going to be able to run laps around people that don't know how to use AI. And I feel, you know, I feel bad because, like, not everyone can learn to use AI. It is, you know, sometimes can be hard to use, but that's just the fact of the matter. It is that much of an accelerator, like it's a 10x to 100x accelerator for productivity in certain domains.
Speaker 2:Do you think it really can be that hard to use? I feel like I saw so. Do you know Allie K Miller? She's an AI thought leader.
Speaker 1:I think I know this name but I don't know that I know her stuff like too closely.
Speaker 2:Yeah. So she writes a lot about it and has a course for how to use AI and whatnot. But she said, if you're confused by how to use something AI related, blame the creator of the AI platform and not yourself. And basically saying that, with the right UX UI, that any AI interface should be easy for somebody to use, even if they don't inherently understand the technical side of AI. And so I would say for anyone who's out there who's wondering, am I someone who can learn it? The answer is yes, and sure it's some trial and error. You know we've talked a lot about before, about the right prompts, about interacting with the information that you get from any type of AI output, but I think anyone can learn how to use it if they feel motivated A friend of mine who launched a product last week.
Speaker 1:They basically were self-taught AI and learned how to use Cloud and GPT to actually write a bunch of code, even though they're a non-technical person, and so I think anyone with the willingness it is at the point now where, if you're willing to put in a little bit of time and effort to learn AI, you can absolutely find a way to take advantage of it. There's enough simplicity and ease of use in the apps that are out there now and it will only get easier from here on out.
Speaker 2:For sure. Okay, where do you want to start?
Speaker 1:Well, I would love to do a shameless plug, if that's okay, to talk about the launch. I'm very excited and it went very well. So you mentioned learning the right prompts, and it's funny because I've been holding off talking about what we've been building until we did the launch, but we've been talking about a lot of the top topics that my co-founder and I have been researching, and this is what we launched a product around. So we launched a. To do is to do two things. One is take away the tedious amount of prompt engineering that primarily AI developers are having to do today. But, you know, pretty soon in the future, I hope, business users will also be able to use our product. So the way we do that is, we are an automated prompt engineering and optimization platform, and so what that means is, if you have a task that you are attempting to write a prompt for, you can actually give us a training set of what we call input output pairs, so basically the input that the LLM would get and then the output that you're hoping that it will output, so the expected output you would like to see, and we use those that training data to automatically generate a prompt for you so it can actually write the prompt for you completely from scratch, if you don't even have a prompt to start with, or if you already have a prompt that you're trying to optimize, it can actually optimize that prompt for you, for accuracy, for the outputs you're looking for. So that's the first thing we're doing, so that's the automated prompt engineering part of it. And then there's this optimization angle, which is either optimizing for accuracy, like I just mentioned, or the other value prop that we've found really interesting lately is, you know, there's a new model that comes out every single week. It's actually pretty crazy, actually literally happened this week.
Speaker 1:Grok 2 dropped this week from X and a lot of developers are constantly asking themselves should I be using this new model, right? Is it going to be cheaper, is it going to be faster, is it going to give me better answers? And so we've kind of coined this term that we're calling model anxiety, which is kind of a play on electric vehicles, so EV range anxiety, where people are worried about running out of battery when they're driving, and so similarly, I think you know developers are facing this kind of constant model anxiety of should I be using this new model, is it going to be better for my application. Is it going to be better for my future? Am I a bad developer? Like you know, a lot of developers have this kind of identification with being the best developer. And being the best developer means you're on the cutting edge, and so am I not a good enough developer? Am I not the best developer from not using the right model? Am I not a cool kid?
Speaker 2:Well, just to add one thing to that which we talked about last time, which is also how much money can you save by optimizing your model?
Speaker 1:It's crazy If you move from GPT-4 or even 4.0, 4.0 mini is pretty cheap, but if you move from 4.0 to Lama 70B, you literally save 95% of your cost.
Speaker 2:Wow.
Speaker 1:That's crazy. It goes down by 95% and you keep your accuracy the same, because what we do is our platform can optimize the prompt such that it will run with 100% accuracy on Lama 70B or 8B, will run with 100% accuracy on LAMA 70B or 8B, as an example. So when you move to smaller models, you have to give it more explicit instructions, because the smaller models have lower reasoning capabilities, and so our platform can automatically do that to help people migrate from GPT to Anthropic or to LAMA, and then also from larger models to smaller models for cost reasons, latency improvements for better UX, unlocking, real-time use cases in their app, things like that. And so we had a really good launch. We've gotten a lot of really good demand coming in both from, honestly, ai app developers and also from actually a few enterprises which are really, really neat and I won't kind of mention who they are publicly, but had some really cool conversations this week. So, yeah, so that was really fun.
Speaker 2:Amazing. I'm really excited about what you're doing. I see the need for it and it's no surprise that the launch went really well. Thank you.
Speaker 1:And I thought one topic that is related to our launch that is really cool is like just the way that I believe entrepreneurship is totally changing with AI, and just two examples of that are. One thing that went viral during our two week break is a tweet about a development platform, an IDE called Cursor, and IDE is where developers are writing code, and so Cursor is like an AI native IDE. So what it means is when some developer is sitting at their keyboard, and so Cursor is like an AI native IDE, and so what it means is when some developer is sitting at their keyboard writing code, cursor is like their clippy it pops in and writes code for them. They can ask questions about their code base and do all these things.
Speaker 1:And I think it was a startup guy, cto or a founder wrote a tweet that got like I don't know hundreds of thousands of views and likes about any CTO or any startup not using Cursor and not using AI to write their code just can't keep up, like you mentioned earlier, like if you're not using it at this point, you're just so far behind.
Speaker 1:Because what he was saying is you get these 10x productivity improvements by you know you can write code in a weekend, that would have taken, I don't know, a month or a few weeks to write, if you know how to use Cursor correctly to write your code. And so it's crazy because you know we think about the whole venture capital landscape and how much money is raised to start these companies, and a huge part of the burn rate in fact the majority of the burn rate of any startup in its infancy is people cost and primarily, engineering cost, and so it really makes me think not going to need to raise nearly as much investment as they have in the past, because we can do so much more with so much less funding. So I think that's really going to change. It'll be interesting to see what happens to funding over the next few years.
Speaker 2:Which is ironic because these days VCs are trying to pile in on AI and very little other things. So the money's there, but maybe the need isn't.
Speaker 1:Yeah, I guess in some ways, especially with the really big funding rounds, the money is going to infrastructure costs as opposed to people costs, Like when you see OpenAI and they raise like a billion dollars, it's because it costs them, like you know, a hundred million dollars or more to train GPT-5. So I guess there is an added cost there. But for us, like SaaS businesses that are just, you know, running out infrastructure, our AI costs are, I guess, high. But that's the problem that actually Nero is designed to solve, which is to move people off of high cost models onto low cost models. So, yeah, I think it's funny that you're right. It's like there's so much money being poured in but there's just not going to need to be as much money spent to create a new company, and I think, likewise, what's interesting is, yeah, as you can see, my mind is all over the place today because I'm so excited about I just think there's so much going on Like from a competition standpoint.
Speaker 1:Let's say, someone launches, you know, Twitter. Right, If somebody launched Twitter today, AI could copy Twitter in 24 hours Like a single developer with, with Claude you know, Sonnet 3.5, which is, you know, everyone's losing their minds over its coding abilities could copy Twitter in 24 hours. And I guess to some degree that was always true. Like it wasn't, it was about the network effects. It wasn't about how technologically advanced the app was, but copycats are going to be so ubiquitous now. It's so easy to do. The bar is so low that anything you create will be copied very quickly, much faster than before, and so I think that's also going to be really interesting for startups to see if competitive dynamics change quite a bit for startups.
Speaker 2:So that does bring me to a few different questions that I have. One is just thinking in general, within the AI sphere, who is easily replicable and who's not? And the first thing that comes to mind are companies that are using AI-enabled hardware, so robotics and deep tech, things like that are probably less easily replicable. And to your point about the costs involved with infrastructure, these companies and then two, I'm wondering. So who is your target customer?
Speaker 1:Yeah, so I think to your first point.
Speaker 1:I'm wondering so who is your target customer? Yeah, so I think to your first point, your first question. I think you're right. I think hardware-based businesses, things that have really deep defensibility, those have always been the quote-unquote best businesses, most defensible business, because they're hard. Hardware is notoriously hard to scale, really easy to prototype, really hard to bring to market like en masse. So I think that's true and I think software businesses are very vulnerable, right, and I think a really good example of that is OpenAI is the most innovative company in the past let's call it 20 years and even they are very susceptible to being copied. Case in point Anthropic Mistral, lama being open source, et cetera, and so even the most innovative company of our generation was pretty quickly and easily copied, which is scary to think about.
Speaker 1:And then to your second point, or your second question who's our target customer? And then to your second point or your second question who's our target customer? So our target customer is AI developers, ai companies who are building products and features on top of these LLMs, and the reason that our target customers is they're the ones who are doing the majority of the prompting right now, and the reality is. There are a lot of business people writing prompts, and that's an area we do want to explore down the line.
Speaker 1:But in terms of the people that have enough of a pain point to pay for a solution, it really is the you know, often venture backed AI companies who either have a AI native product, so they're starting from scratch and their whole product is built on LLMs, or larger companies who have significant features in their product that are is built on LLMs, or larger companies who have significant features in their product that are based on these LLMs, and so they're the ones that have the pain point. You know the budget, the need to use something like us now and then. What we want to do is expand later down the line to more business users, right Like the everyday person who's trying to use an LLM chat GPT in their job. They also could benefit from help with their prompting, and so, down the line, I think we'd like to release a product along those lines as well.
Speaker 2:So for the first phase, right now, are there any names? It doesn't have to be someone you're actually working with, but any names of companies that people might know to understand. Oh, this company would be a good user.
Speaker 1:Oh, that's a great question. I mean, you know, I think for our listener base, everyone's got a friend who's probably working on an AI app these days, right, somebody who has launched an AI app. I'm trying to think of. We have some customers who it's kind of funny. Some of our customers see having us in their product as a competitive differentiator. So they were like, hey, can you hold off on like saying our name yet until we get a little bit further ahead? So that's been kind of funny. I'm trying to think of some of the ones that we had conversations with this week. So here's an example Like we had a conversation this week.
Speaker 1:I actually love this. It's a mental health app, so I won't name them because we're still obviously in conversations with them, but mental health app, and so it's AI assisted therapy, and so it's an app that you can download and it has these really neat pathways for if you're having anxiety or if you're having depression, and it can kind of send you down these CBT based chat bots that are based on, you know, psychologists and you know, help you kind of, you know, deal with your mental health needs Right, which I think is a really important topic. I think that's really like a great app to have out there. I think AI is going to change the world of psychology and therapy for being always present, always available. You know, if you have a therapy appointment, maybe your appointment's not for a month or a week and you need something more in the moment, and so you know they want to make sure that they can serve their customers the highest quality interactions right. So it's therapy.
Speaker 1:It's very, very important that the interactions are very high quality, and they also want to make sure that you know for them as a business, it's cost efficient, right, because it's no good if you know it's so expensive to run these high reasoning models that they're not able to actually offer this platform at an efficient cost for their customers, right? And so you know those types of apps where you know AI therapist, or maybe an AI productivity app is another one we're talking to this week. So there's a really cool AI productivity app out there. Actually, I'll call this one out. This one's called Hoop, because it's our friend, stella I know Hoop.
Speaker 2:What they're doing is super cool.
Speaker 1:Hoop is very cool. I've just onboarded as a user this week and so we're talking to them about trying to help them out with some of their prompting and cost efficiency as well, because one thing that's really interesting about AI costs is I think we've all gotten used to freemium services. You know, trying out this new consumer SaaS product and it's free to use maybe for two weeks or, honestly, a lot of them are free forever, you know, for a very basic type of subscription. A lot of them are free forever, you know, for a very basic type of subscription. But that business model is very hard to do with the current cost of AI, because AI is much more expensive than traditional, just like cloud computing, and so this like free forever concept is very expensive actually for AI apps, whereas it wasn't expensive for traditional SaaS apps, and so those are the types of apps that we're kind of working with. You know AI productivity apps, ai therapy apps anything that uses AI can have a use for us.
Speaker 2:That makes a lot of sense. And what you just said about the freemium model I noticed with ChatGPT. The free version has a limit of how much. I'm not sure exactly how they calibrate that limit, but how much you can use it until it then forces you to upgrade to the paid version, and so I do think you are seeing more of that. I pay for the versions that I use, so I haven't noticed that, but in talking to other people who have just been experimenting with the free versions, they said they did get capped out.
Speaker 1:Yeah, exactly, and what you'll see with AI apps is it's kind of interesting. My co-founder noticed this apps that you normally would have seen like a 30-day free trial period. Now you'll see like a one-week free trial period and you're giving your user way less time to get accustomed to the product. But you have to do that because it's so much more expensive to service users with these AI products because of the cost of LLMs, right, and so in that way, hopefully we can help bring down the cost for some of these apps and then they can offer those 30-day free trials or maybe a free forever type of plan, because it's much more cost efficient.
Speaker 2:Interestingly, I've noticed just a new trial model in general. So, yes, time is one thing that I've noticed has been reduced. So, yes, now the seven days are 14 days instead of 30. But I've also noticed it's now with more usage. So you get a number of free credits to try, or a number of minutes if it's video production or something like that, and then after that it forces the upgrade. So that's also interesting to think that just the concept of a free trial has really been flipped on its head.
Speaker 1:Yeah, 100%. And it's funny because those credits are 100% directly correlated to the token cost on the backend, right? So some number of credits, they give you, some number of video analyzed minutes or some number of blog posts you can generate for free. What they're doing on the back end is they're saying, okay, the average blog post is 600 words. 600 words equals X number of tokens. Then X number of tokens on GPT-4 will cost us Y dollars, and we're willing to eat Y dollars on every user per month or whatever. So that's the math. That's dollars on every user per month or whatever, right? So that's the math that's happening on the back.
Speaker 2:Mm-hmm. Yeah, it's a new way for them to think about customer acquisition cost.
Speaker 1:Exactly right.
Speaker 2:Okay, so also another update. This is how quickly things happen in this world. Another update that happened since we last recorded was speaking of Llama. I think it was like 405B release, so all within a few weeks. How much has come out is just really crazy.
Speaker 2:But let's go to this image generator, because I saw I think it was an evolving AI on Instagram. It was a side-by-side one photo of a woman speaking at a conference next to a photo of another woman speaking at a conference next to a photo of another woman speaking at a conference different background, different people and it said which one is the AI-generated one. And I stared at this photo for so long and looked at all the details that you normally see are slightly off with AI. So maybe it's the text, maybe it's the fingers, maybe it's the teeth, and for the life of me I had no idea. And then I decided OK, I think it's the right side image, because the background's a bit blurrier and the lighting's not as good and there was a weird shadow on the mic that maybe AI wouldn't be as good at realizing that it was the wrong lighting.
Speaker 2:And then the next day it gets released. No, it was the left side photo. That was AI generated. That looked so realistic. It had a Google badge, a TED Talk background, it was nuts, and actually it's a whole video that's AI generated, and so to see that level of detail is mind boggling. And then that brings us to 0.1 and run with it. Tell us what you're hearing on Twitter or X.
Speaker 1:Twitter is a fire, a flame, I don't know Whatever the word is. Twitter is going bananas. A light, a light. Flix. Twitter is a light with Flux. Yeah, the Flux model. It's absolutely crazy. I spent probably too much time playing with it today, if I'm being honest. It is so addictive. So Flux is the new image model. It is a competitor to MidJourney was previously the kind of most well-known image generation model.
Speaker 1:We call these text-to-image models, which means you type in a text prompt what you'd like to see and it will produce an image. And Type in a text prompt what you'd like to see and it will produce an image. And then, yeah, famously, the one example that's going around is a woman kind of giving a TED style talk. So being on a stage giving a kind of a tech style presentation, it's unbelievable. I mean, the realism is really astounding. I was going to say the realism is unreal, no pun intended, and yeah, that's really funny. I didn't know we hadn't talked about that, that you got it wrong, and I don't blame you for getting it wrong, because how could you get it right? Like it's hyper-realistic, it is realistic.
Speaker 2:All the comments. Got it wrong too, by the way.
Speaker 1:Really yeah, and so what people are doing is to take it the next step, which is really cool. People are taking Flux, generating a hyper-realistic image, and then they go to another model which is called an image-to-video model, and there are two big ones they're using. One is called Runway Gen 3. So you take Flux, you generate a cool image and then you pump it into Runway or Kling and then you make that into a video.
Speaker 2:And I've been having a lot of fun with that tonight, and I feel like it's a great lead into what we first started talking about today, which is how can you have an entire launch of a new product with as few people involved as possible. And so why don't we do the side-by-side of let's call it, the old way of doing a product launch and the new way, using AI?
Speaker 1:Yeah, that sounds great. Where do you want to start? Do you want to like prompt on certain departments, or do you want me to like talk about how it would look?
Speaker 2:Let's start with the old way. Who would have needed to be involved with a product launch?
Speaker 1:Yeah, so I think it starts with, I mean, just building the product in the first place, right? So initially, or not initially, I mean the old way is you raise, like a pre-seed round, or you raise a seed round and you hire, you know, a small development team, so you usually have two to four developers and you build out your product, right, and so that would have been the old way. New way, you know, maybe raise money, maybe you don't, and then you know one developer is doing the job of three, right? So instead of hiring either some low-cost offshore developers or hiring some, you know, significantly expensive developers, you've got one cto who can make use of three I was gonna say three agents. Actually, I'll take this even further and say so.
Speaker 1:The other company I wanted to call out today my friends launched a company. It's called Soft Landings Meditations. It's a very cool company. I'm very on board with the mission of this.
Speaker 1:It's a woman who had been writing what she called meditations, which are basically these kind of audio stories. I think they're about five or six minutes in length, and she was writing them for her kids and recording them for her kids, to help her kids with really common issues that kids have between the ages. Right now they're focused on between the ages of five and nine, so it's things like wetting the bed, going to bed alone, going to sleep with the lights off, getting car sick, things like this, and so in her spare time she was, I think, recording these herself and has now turned to AI to help her generate some of these things, and they're highly personalized and customized to the children, which is really cool. So like if you have a nickname for your grandma. I think they had an example of like oh, someone's kid calls their grandma like Gam Gam. You know, when it's talking to these meditations these kids listen to when they're going to bed, it'll be like, oh, gamgam is so proud of you, and like that personalization really matters for these kids.
Speaker 1:And they've really found that kids love listening to these things and actually are requesting more of these meditations. And so, like my friend was telling me a story today where the you know, one of their customers has a kid who was using this meditation for sleep you know, being able to sleep alone and requested another medication for car sickness because they like the character so much and they like the person.
Speaker 2:Wait, you said medication, but you meant meditation.
Speaker 1:So it's a meditation. It's an audio meditation.
Speaker 2:But it sounds like it is a medication, a non-drug medication.
Speaker 1:Yeah, this is like audible or you know, like calm or headspace, right, it's like these. You know these guided audio kind of recordings. And they requested one for car sickness because they love one developer. But actually, in their case, the old way was hiring a development firm. I think they got quoted $50,000 for their website and their app and all this stuff for their prototype.
Speaker 1:And Julie, the founder of this. She literally figured out in her free time how to use Cloud to create these multiple agents, which is actually a very advanced technique, and so she would say, hey, cloud, I want you to play three different roles. I want you to be an engineering manager, I want you to be like a designer and I want you to be an engineer. And here's the app I would like you to build. Can you figure out the plan for building this? Can you execute on the plan and can you design this app? And she showed it to me last week and literally executed on a website and an app. She is not technical. She does not know how to write code. She did it all with Claude and she did it successfully. I mean, that's the new way. It's like not even having a CTO. Like Claude is your CTO and that's. It was honestly mind-blowing.
Speaker 2:That's incredible. I'm guessing she must have some technical background, if not on the engineering side or ML. At least you know something that gave her the ability to parse through that information and get it to that point. But that's still super impressive.
Speaker 1:You know, honestly, you would think that I'm not sure that she did to be honest. So she she worked at IDEO.
Speaker 2:So I think that that is part of it. I mean sure, OK, so she worked at Innovation Design yeah.
Speaker 1:So she was in the world of understanding kind of what apps like look like and feel like, but she was not.
Speaker 2:And then she had worked with engineers but was not an engineer herself, and so what's cool is but that's a perfect use case for somebody yeah, Like you yourself are an engineer I'm not, obviously but for somebody like me who has been maybe the translator between the technical side and the market or on business development or anything where you're sitting in between those two so you know just enough to understand what should be happening, even if you can't execute it yourself. And then you have this tool where now you can execute it yourself, and that's really incredible.
Speaker 1:Yeah, it was crazy. She was like she was like plan this out. And it said, here's our stack. Our stack is going to be, you know, JavaScript, it's going to be React, it's going to be all this stuff. And she literally was like, what does that mean? And it would say, oh well, this is for the front end and this is for the back end. And she'd say, okay, well, what do I people on this podcast to do is like just keep asking all the questions. And don't get me wrong, this has taken her hours and hours and hours of effort and she has some free time. She's like in between, I think she's like in between some jobs, or she like recently left her job and so she had some free time to like work on this. But it is possible and it's only going to get easier from here on out. And so I'm just really blown away by.
Speaker 1:I think about all my friends, you know, having, you know, formerly been an engineer. How many people have come up to me like, oh, you're a founder, I've got this million dollar idea. Like people do this stuff all the time and I love it. I actually think it's really fun to talk about their ideas with them, but no one can ever execute on it, because entrepreneurship traditionally has been about execution and still will continue to be about execution. But people say I got a million dollar idea. It's like all right, why don't you go do it? Well, I don't know how to build it or I can't afford to pay someone to build it, and now you're going to be able to just like go on cloud and say, build me this app, Right, and so that's just. We've just covered just the development piece and we haven't even talked about, you know, the old way and the new way for marketing design. So yeah, we can go into that as well, if you're.
Speaker 2:Yeah, let's go there.
Speaker 1:Yeah, I mean, marketing is a big one for us. You know, I did all the marketing for our launch using Claude, and so I put Claude in the role of a product marketer and I said, you know, I had this really elaborate, long, prompt and I said here's the app we're building. We're building an app for AI developers. It helps optimize for, you know, cost and latency and accuracy for LLMs. And I gave it a bunch of descriptions about what we're doing, what's our value prop, who's our ICP, et cetera. And then I said generate some website copy, you know. Generate a hero image. Generate a copy. Generate three value props with associated detail copy. Generate call to actions. And it did all of that and it was really incredible.
Speaker 1:And what I actually did was I wanted to evaluate a few different kind of positionings, and so what I mean by that is, you know, one positioning I wanted to experiment with was personifying our app, right? So a lot of people will say, like these apps have like names, like Claude, right, or something. So I wanted to experiment with one positioning that was a personification of our app. I wanted to experiment with one positioning that was our app was positioned as a copilot, so kind of like GitHubpilot, which helps people code. It's a prompting co-pilot. And then I wanted to do one more like kind of traditional style developer tool, positioning Right.
Speaker 1:And so pretty quickly, within a couple hours, I had actually five candidate positionings and I was able to show them to people and say, like, what do you think about this? What do you think about this? That would have taken, you know, a copywriter, a marketing team. You know that would have taken, you know, a copywriter, a marketing team. You know that would have taken a number of people like a week to show around. I think in the past a founder would have been like hey, like here's my marketing consultant, could you give me five different positionings? They would have gone and done a bunch of research, they would have written a bunch of copy. They would have come back to you a few times with, like what do you think of this? Et cetera. I did it in two hours. So it's amazing, I think things that usually would have taken whole marketing teams and, you know, a whole week of turnaround is just like trivial now, which is unbelievable.
Speaker 2:I feel like I'm just catching up to the enormity of this right now, as we're talking about it and thinking how much one person is now capable of using AI is astounding in and of itself. And then two this idea of specialization, and will that be necessary in the next five to 10 years? So instead of saying I want to go into marketing, do you say I want to go into what?
Speaker 1:Yeah, I don't know, it's a really good, I think. And I mean, maybe a less extreme example might be in marketing. There's so much specialization, right. You've got content marketers, you've got product marketers, you've got brand marketers, you've got designers, you've got all these different roles, and now you can get these AI agents to play these different roles. You could say, hey, you're the brand designer, you're the product marketer, right, and you can get them to play off each other and interact with each other the same way a human team does, and so Can you talk to?
Speaker 1:me about that for a minute.
Speaker 2:I want people to really understand this because I did this exercise with someone recently. I'm helping them launch something and showing them how they can use ChatGPT or QuadRoc for it, and I said to it tell them or write in the prompt you know as a this position, walk through this. And they said, really. I said yeah, like tell it who you want it to be in this conversation and I even something to us that seems simple like that. I think a lot of people don't understand how they can really unlock that and so, just if you could talk through that for a bit, yeah, I think that's a good point.
Speaker 1:Sometimes I get ahead of myself and forget that some people don't understand some of these techniques that have been adopted, so I'll break it down like really simple terms. So you're sitting in ChatGPT and there's a blank box in front of you and you give it a prompt, you ask it a question or give it a task or whatever it is. Do you remember? Actually? Let me ask you one question Does ChatGPT have a system and a user prompt, or does it just have the blank box? Do you remember if it shows you, hold on, let's see, let's just skip this a minute, hold on.
Speaker 2:Of course I have ChatG. Yes, I have it open. So when you say open a system prompt, what do you mean by that?
Speaker 1:Is there like a settings button somewhere on ChatGPT?
Speaker 2:There is no.
Speaker 1:It usually would be like in the left sidebar or across the top somewhere. This shows you. It's funny. I do everything through like the developer workbenches.
Speaker 2:Funny enough, my CTO kind of does everything through the consumer version of ChatGPT, which is, oh, that is funny, okay. So in the upper right-hand corner is what you see on a lot of websites where it's like your account with your little icon Got it, and if I click on that it says my plan, my GPTs, customized ChatGPT settings.
Speaker 1:Okay, so when you do customized ChatG, chat, gpt, what does it say there?
Speaker 2:custom instructions. What would you like chat gpt to know about you? To provide better responses? How would you like chat gpt to respond? And then I can pick browsing the dolly image creator or code and then I can say enable for new chats toggle, toggle on or off.
Speaker 1:So I think that the custom instructions there are what we would call a system prompt. I'm not 100% sure about that, but in the developer version of GPT you can give it a system prompt and a user prompt and they're largely the same. There is a small distinction, but basically the system prompt is preparing the background instructions or persona of the chatbot, if you will right. So, like in the system prompt, if you say you are a product marketer, you're an expert technical marketer at you know b2b, sass applications, then what it will do is everything you ask it from then on will use that context before it answers. So, and I think what they're doing on the back end.
Speaker 1:I could be wrong, but basically the system prompt is one way of saying every time I ask a question, I'm actually going to append the system prompt to what you're asking so that it answers through the lens or through the context of the instructions that you've given it, the persona that you've given it.
Speaker 1:And so one way that a lot of people I mentioned this because one way that a lot of people give these personas to their your point earlier they should do exactly what you mentioned, which is in the blank box that you have before you ask it the task, you often will preface it, or you should preface it by saying as a product marketer or as a brand designer or as an expert in whatever the task is you're about to ask it, you give it a personality and I'll expand on that by saying don't just say hey, as a product marketer, answer this.
Speaker 1:You would say something pretty detailed, like as a technical product marketer who's an expert in B2B SaaS marketing for developer tools, please complete the following task. And in doing that you actually influence the answers that are going to come out, because I won't get into the math of it essentially, but basically you're like pre-training the model to understand the context in which it's operating, and so it knows kind of where to search in its universe of knowledge for the answer that it's looking for, and so you will get a more correct answer or a higher quality answer when you give it that type of context to work with.
Speaker 2:Yes, and I think it's a really key point because I've told people as much as the models are trained on the vast data of the internet and then some, as you use it, you can train it for your own purposes. So as you give it prompts like that, it gets trained on what you want it to do within the context of the question or whatnot. So I think that is really important for people to understand and to remember when they're using any llm. It's not specific to chat gpt that you can get that detailed with it and the value I mean this is why you have the business you have the value of a really good prompt in making in, then giving you really good output.
Speaker 1:Yeah, and what's cool is one thing that Julie did. She experimented with giving a different persona to multiple chat, what we call agents, and so sometimes people use this word agentic, or more likely you'll just hear the word AI agent, and what that means is an agent is essentially a chatbot with a defined persona, and so if you open up ChatGPT and you give it, you start with the system instruction or you just start with the instruction of saying you are a software developer. You've now created the instruction of saying you are a software developer. You've now created an agent. That agent now is a software developer agent, and so what some people are doing is they've got one chat GPT window open over here and they say like you're a software developer, and they've got another window open over here and they say you're a marketer, and they will like copy and paste the answers among them to create this type of interaction between agents, just like a software developer might talk to a marketer in real life.
Speaker 1:Right, and what I thought was cool about what Julie did is she experimented with having multiple agents that were separate and copying and pasting answers across, and she actually found she got much better responses when she just gave one one chat three personas and this only works for, like, higher reasoning models, and so she was doing this on claude sonnet 3.5, which is, you know, one of the higher reasoning models, but she actually just said in there, like I mentioned earlier, she gave it a prompt that said you are a team of a designer, an, an engineer and a product manager. And what was really funny about this is she showed me her chat history and it was unbelievable, because what Claude did was it took her task, which was design this app for kids' meditations, and it created conversations between the multiple personas. It would be like engineer says this I should do what would you like me to do? And it would be like product manager says you need to do this, and it would have these whole elaborate. It was like sitting in a team meeting. It was crazy.
Speaker 2:It's really amazing. I feel like you have to give them even extra personality traits, like you're a contrarian or you're collaborative, or you are always playing devil's advocate, or something like that.
Speaker 1:Totally and it will actually do those things. And again you have to use the higher reasoning models. But I think one of the funniest examples was at some point and she showed this, it's so funny At some point it said the engineer said, okay, great, I know what we need to do. It'll take me about two weeks to do this. And she just messaged back Great, that sounds really good. It's been two weeks, How's it going? And it came back and said it's going, great, I'm done with it. It's unbelievable. I mean, this is truly a new world of entrepreneurship and company building because, yeah, this non-technical person spun up a whole team and was able to reproduce something that was a $50,000 quote from a software like an outsourced software development firm Unbelievable.
Speaker 2:And honestly, all things considered you know, having gone through those types of processes myself and for you too all things considered, I don't think she probably spent that much more time doing it herself than she would have going back and forth with the company.
Speaker 1:Yeah, I agree with that and like obviously there's some learning curve to it, but the quality is pretty good. You know, like the quality of code that gets written is pretty good, the quality of the marketing, the product marketing stuff that I was doing, the website copy, etc. Pretty good. You know, every once in a while we change something, but I find the quality to be pretty on par with what we'd see in the workplace oftentimes.
Speaker 2:Definitely, and every week that goes by it gets better and better. So now that she has this to start a month from now, she can take that and improve it even more and maybe beyond what she would have gotten from that $50,000 project.
Speaker 1:Because that's a lot more static.
Speaker 1:Yeah, and like you know, honestly I hate to say it, but you know I've worked with some outsourced firms. Some are amazing. Right, we used a design firm. You know I will say we used a firm called Klimpton Design. I will give them a little shout out Klimt, a-o-i-t. Klimt and Design, incredible. So we use them. So we do use some outsourced firms and they're incredible.
Speaker 1:And that was for some design stuff that I think is harder to do in the image models right now. Like, I think you can do some basic logo stuff, but like, when you're looking for really overall brand design, I still think that's really good outsource. But when it comes like outsourcing, like product development to you know some firm wherever in the world, there's communication barriers, there's time zone barriers. It's genuinely easy to communicate, easier to communicate with these LLMs than it is to schedule a meeting at some weird time. You know, when you're going to bed to talk to the developers on the other side of the world and you can iterate so much faster. If it doesn't do the right thing for you, you can just be like, okay, cool, no, you messed that up, do something else, whereas you know, with these offshore development firms, you got to wait the 24 hours to see what they did the next day. So it just moves so much quicker.
Speaker 2:It is really amazing. It's as soon as you want something from it, it's ready. It's ready whenever it's ready, whenever you are, you know. So, yeah, it's really amazing. I'm wondering, best guess, how much money do you think you saved in this launch doing it the way you did it, compared to what would have been the old way?
Speaker 1:I mean I will say this for us, because obviously we're like you know we have some funding but you're trying to be scrappy we saved a lot more time than money. I would say, you know, if I wanted to outsource this stuff, I would say in the tens of thousands, because if I was going to outsource this to some marketing consultant, you know that would have been, you know, 5k or something like that. And so I would say it's definitely in that. Developers, you know we're using, obviously, cloudsign at 3.5 to write a bunch of code. So that would have been a developer. So I mean it's in the tens of thousands of dollars. But for us, since we would have just done it ourselves to save the money, it would have taken us, you know, much longer to launch. It would have taken us much longer to write the copy. It would have it's much longer to write the code. And so, yeah, it saved us a lot of time.
Speaker 2:What's much longer A week, a month, six months.
Speaker 1:Yeah, probably on the order of weeks to months. I mean, I would say this was definitely something. When I wrote the copy in two hours it would have taken me previously, not to say it would have taken me a week. But there's just a lot of stuff that goes into writing copy. You're researching what other copy looks like. You're trying to be original, you know. You're like, okay, now I've seen this thing, my brain is primed to sound like that thing, so I want to sound like that thing, but slightly different, and that stuff's all you know.
Speaker 2:that's like it's laborious, right, it's tedious. So I would say it was a four to five X accelerator for us, if not like a 10 X accelerator in certain areas. It's incredible, and then I mean if you think about opportunity costs and costs of entering the market three months from now versus today. Obviously we don't know the exact dollar amount of that, but there's definitely some dollar associated with that.
Speaker 1:Yeah, and also, you know, just from a quality standpoint, you know previously where I might've said oh, let's do three iterations of this Cause it's like pretty tedious. Now we can do like, oh, let's do five different versions of this and look at it Right, so you can actually generate more possibilities and hopefully end up with a higher quality product. I do think there's like some curve here where it's like okay, well, I can, can generate 100 of these, and then it starts to become too much noise, and so when the bar becomes so low the barrier is so low to do stuff like this you start to get a little bit of noise. But I do think, yeah, you can either. You can kind of go whatever direction you want, you can save money, you can do it faster, you can get a higher quality, depending on what you're optimizing for.
Speaker 2:I guess my ending note here would be to say again where we started, which was not losing jobs to AI, but losing jobs to people who know how to leverage AI.
Speaker 2:And I think the same in this type of situation where it's not to say you're going to do everything yourself for every product launch out there.
Speaker 2:At the end of the day, there are still many reasons why you would want to partner externally or bring in other people to do certain things, but because of the resources that we have with AI, now you can be a lot more selective about who your partners are and really targeted and focused on why you need them and what they'll do. And so, from the marketing side of things like the outsource design that you have, I think that still makes a lot of sense, because some of these image generators and some of the things that we've been talking about, yes, they are available to consumers, but at a much lower quality than at the enterprise level, and so these design firms that are using those AI tools for enterprise can do so much more right, Because then they pack a double punch of understanding design and knowing exactly what they're trying to do and having access to better quality resources. So that's how I see it overall, but curious to hear your thoughts on that.
Speaker 1:Yeah, no, I think that's exactly right. I think you nailed it. I mean, I think that you know everybody the best of breed in every discipline, is going to be using AI as an accelerator in some form or fashion to get their job done. It's that the most creative people are going to learn how to use AI, know how to use AI to increase their creativity.
Speaker 1:To get more produced with their creativity. And so you know, there are obviously some jobs, some roles that will get replaced by AI, but I think for many of them, AI will be an accelerator and it'll be additive and it'll make us all better our craft out.
Speaker 2:I think so too. I mean, I know myself and I'm no design guru, and just seeing what designers come back with when I give them an idea or concept is something that my brain would have never thought about. And so, even using AI, could I get that optimal output? I'm not sure. I still would want to have someone else in that process who really knows it, but I'll say for sure I don't want to work with someone who's not using those tools to make it better.
Speaker 1:Yeah, I think that's very well said.
Speaker 2:Awesome. Thank you so much for sharing all those details about Nero. I'm really excited. I know this is just the beginning of huge growth and a really you know I guess I already said excited but a really exciting journey and I'm sure, week by week, we'll get the updates.
Speaker 1:Yeah, I will keep you updated. Shameless plug before we end. So getneroai is our app Also. The other ones I mentioned because I love and I want you guys to check them out too is Soft Landings Meditations. You can find them on their website if you Google it or you can find them on Instagram. And the last one is Hoop, which is our friend Stella, who has a really awesome productivity app that's based on AI and I've been using this week and it's really fun to use. So go check out Hoop app. I forget their website.
Speaker 1:It's like get Hoop or and Klimt, klimt and Design.
Speaker 2:We'll put them all in the resources in the episode resources. Awesome Thanks, Paul. Talk to you soon.
Speaker 1:All right, have a good weekend.
Speaker 2:Okay, bye.
