June 07, 2024


Prompt Engineering

Hosted by

Jordan Gal Brian Casel
Prompt Engineering
Bootstrapped Web
Prompt Engineering

Jun 07 2024 | 00:59:28


Show Notes

Launch week.  AI demos.  Prompt engineering.  Help docs.  Product marketing.  New feature prep.  B2B vs. B2C strategies.  Coastal living.  Mainstream AI.

Connect with Brian:

Connect with Jordan:

View Full Transcript

Episode Transcript

[00:00:17] Speaker A: Hey, it's bootstrap Webb. It is Friday, June 7, and we're back at it. Today is the last day of school for my girls, and, yeah, summer is here. Let's go, Jordan. [00:00:28] Speaker B: Yesterday was the last day of school for my girls, and today is already complete chaos. My oldest had four friends sleepover. The middle had one friend sleepover. Then more friends came over at, like, 08:00 a.m. there's air hockey happening. The boys came to visit at, like, 09:00 a.m. wife went out to get bagels. I mean, here we are. So it's fun. It's good chaos, but summer is here. [00:00:55] Speaker A: That sounds fun, man. [00:00:56] Speaker B: Yeah, it's fun. And this town has a great thing. They do the last day of school, right after school, which ends early, everyone goes to the beach. So the entire town is at the beach. It was lucky. It was, like, 75 degrees and sunny yesterday, and it's just like a fun celebration thing, and we're lucky enough to be ripe on the lake. Yeah, it was like a beach day. I took half the day off. Everyone's having a few drinks. It was beautiful. [00:01:25] Speaker A: That's awesome. So how far are you from the lake? [00:01:28] Speaker B: We are about five blocks. [00:01:30] Speaker A: Oh, yeah? You're real close. [00:01:31] Speaker B: Yeah, it's a trip. It's like, we live in Illinois, but it is a beach town. It is, right? [00:01:37] Speaker A: Yeah, for sure. Yeah. I remember living in Chicago, not far from the lake. And so here in Connecticut, we live, like, a ten minute drive to the Long island sound beach and absolutely love it. Like, you know, it's one of these weekends we're gonna start going. We like to go out there, like, every weekend, do breakfast on the beach, and you grew up in Long island? Like I did. I don't know about you. I feel like I can't live anywhere that is just totally landlocked. And just, like, you think about, like, so much of our country is. But I've always grown up near a coast and having, like, easy access to a beach. [00:02:12] Speaker B: I mean, if you look at the population map, a lot of it is centered around the coasts. You know, that's true. But I'm with you. Portland was an anomaly. We went out there thinking, hey, it's kind of by the beach, but it's a good hour and a half drive to the beach. And the beach is so unpredictable. It could be 75 and sunny in the city, and then you drive out there and it's 52 degrees and raining, or it could be 90 and sunny. It's very, very unpredictable. That was the departure. But we're very happy to be back by water. [00:02:43] Speaker A: And I love Lake Michigan and the Chicago area. I feel like it's actually kind of similar to where I live now in Connecticut, where it's like, you get the benefit of. It's like a legit beach. I mean, Lake Michigan is, like, practically an ocean, but very calm, very calm waters and everything. But the cool thing about it is that depending on where you are, it's not super crowded and it's pretty clean. Like beaches and clean water. And the same idea here in Connecticut. I grew up in Long Island, New York, 15 minutes from the Atlantic Ocean. But it's crazy traffic getting down there, and then it's just insanely crowded on the beach. You've got beach towels, like 2ft from the next person. [00:03:29] Speaker B: It's good when you're 17, bad when you're four. [00:03:32] Speaker A: Yeah. But here, like, up in Connecticut, it's just, like, totally wide open, nobody's around. Super clean, super calm. Love it. We spent so much time down there. [00:03:41] Speaker B: Yeah, yeah. Anyway, all right, now, on the work front, got a lot going on. We're about to launch. It's about to happen. It's real. [00:03:51] Speaker A: Oh, all right. So what does launch mean? What does this look like? [00:03:54] Speaker B: Yes. Okay. Fair. Fair question. And Patrick, you know, gonna disappoint you, not gonna give the URL we are set to publish on either Monday or Tuesday. And the delay, we thought we were gonna publish this week. The delay was two things. One, the demo video that we're trying to put together, that just always takes longer than you think. And we wanna do a good job because that's really the only interaction with the product that we're gonna put out there. We're not gonna just put out a phone number just yet. At some point in the near future, we'll either put a phone number or a button that you can just click and just talk to our agent. We're not quite there yet. [00:04:33] Speaker A: So. Like your AI agent. [00:04:36] Speaker B: Yes. Yeah. It's a weird thing. And rock has a lot of scar tissue from the checkout product in terms of technical paranoia. You know, at rally and at card hook on the checkout, we needed technical paranoia. How are people going to jam in coupon codes? How are they going to use fake credit cards? We just. We had to think in a paranoid, you know, perspective. But this is. This is also tricky, because if you put an AI agent out there, whether it's up on the site or behind a free trial that doesn't require a credit card, you are taking a real risk, because hits to the API means something very different. You know, it's money per minute. And so if someone wants to be a bad actor, you have to protect against them just blowing up your agent. And then the next morning you wake up to a $5,000 bill from OpenAI. [00:05:36] Speaker A: Right? [00:05:37] Speaker B: Yes. [00:05:38] Speaker A: Yeah. And in that case, in your case of like, using it as like a demo to show on your site, you know, with no credit card up front, no sign up up front, you're just gonna get a ton of tire kickers and you're paying for those. [00:05:53] Speaker B: Yes, that's right. And to a degree, tire kickers are. [00:05:56] Speaker A: That's fine. [00:05:57] Speaker B: Good for now. Yes. So there's a balance there. So the video that we're putting together, you know, the way we're doing it is we want it to be a video even though it's an audio recording, but we want to show on the screen what's happening. So there are times when you're just seeing text on the screen and you're seeing the word being spoken as highlighted. And then at other times we want to show the calendar view and how a calendar is read and then appointments are created, and then we want to show like an sms confirmation. So there's some stuff to do there. And the second reason I feel like such an amateur on this front, privacy policy and terms of service. At some point, I was like, oh, we can't use the same one as rally. There's a whole bunch of different issues. It's not about payments, it's about calling and recording and two parties and all this other stuff. So I hit up the lawyer, and that's not an overnight thing. So it took a week to get that done. I just got the email from the lawyer today. [00:06:56] Speaker A: I don't know if I should be, like, broadcasting this, but I've taken the bootstrapper approach to terms of service and privacy policy, which is like, you find whatever template out there and you swap in your company name. And I've used, like, in many cases, like the same one across multiple products of mine, just swap out the company, make some tweaks for some terms that you need, whatever. [00:07:15] Speaker B: But, like, I think that's how almost everyone should do it. I think that's generally the right approach. And I say that to the lawyer up front. I go, I just want you to know, my perspective on this is that no one will ever read it. It will never come into play. This has only negative impact on the business, you know? So I kind of, I make sure the lawyer knows my point of view on it because I know what's going to happen, and it did happen. What's going to happen, especially around phone, is that as soon as you touch legal, they will come back to you with feedback that will be a negative impact on the product and service. And then you have to make a decision on the balance between how strict to be on legal and risk and how much you're willing for that to impact the product, you know, because that sucks. It sucks. At first they told us, you can't have recordings. You need to delete them right away. And we were like, no, absolutely not. [00:08:16] Speaker A: So like, your actual customers? [00:08:18] Speaker B: Yes. [00:08:19] Speaker A: Or, yeah. [00:08:20] Speaker B: If you ask a lawyer, should a service like ours keep recordings of calls? Every lawyer will say, absolutely not. Huge risk. Don't do it. [00:08:30] Speaker A: Absolutely. [00:08:31] Speaker B: And then you start to poke and prod and show examples and then all of a sudden, oh, we've come up with a safe way to do it that allows us to keep the recordings. So I, like, knew it was going to be this funny little battle between someone that I'm paying, but they're helping. [00:08:46] Speaker A: Yeah, yeah. I mean, it is their job to, like, find the areas of risk and, yeah, push back on them, you know? [00:08:52] Speaker B: Yep, yep. [00:08:53] Speaker A: Yeah. Very cool. What was it gonna say? [00:08:56] Speaker B: What do you got going on this week? [00:08:58] Speaker A: You know, this week, like, today feels pretty good because I finished probably three significant projects that were on my plate. These are like, I think of them as like, big blocks of work that take up significant amounts of time. Multiple days, and these are shipped. And I've moved, moved on, and I'm moving on to the next thing. Like right now. Today I'm like, in between those big projects and about to start the next project. So two of those projects are in clarity flow. One of them is a client project. And there's just that mental transition state that it's sort of a love hate relationship. It lasts about a half a day to a day where it's like, okay, that's done. It's a huge check mark on my. I like to think of a Kanban board that's a big card that I've just moved into the done column. [00:09:50] Speaker B: Okay. [00:09:51] Speaker A: But then, like, as soon as I do that, I need a little bit of time. Like, I took off work a little bit earlier this week and I'm, you know, hanging with the girls and doing some hobbies and stuff. And I'm starting to, and it's like, I know exactly what I'm gonna be working on next. But there's still this, like, mental state where it's like, wait, is that the right thing to work on next? But what if I work on this next or, or that next? Okay. And it's a little bit of a decision around like, well wait a minute, like don't, don't get into something because whatever you start, you know, is going to take. It could, it could take several weeks. So, but anyway, like, the two things on clarity flow, one is product related. So we are starting, we're now breaking ground on a very big new feature in clarity flow, something that customers have been asking for, like, like requesting forever. And I've been mostly saying no to the feature and I'm, now I'm cracking, I'm like, no, we should just do it, let's do it. My role in that project is I've gone back and forth on different ways to execute big features in clarity flow. One is I just write up the specs in linear, in detailed specs that I'll give to my developer, but then after she's done, I will build most of the UI component, like clean up the UI components. She'll put in the functionality, but then I'll make it look polished and ready to ship. This time I did all that work upfront, or most of it upfront, like the views, the components. There's a lot of new interface stuff that we're going to need for this new feature and I put a lot of that in place and I wrote multiple issues in linear super detailed. I'm basically queuing her up for like at least five or six weeks of her full time focus on this feature. So I spent about three, four days on that stretch of work and now the ball's in her court. She'll be working on that for the next month or so. And the other one is sort of a marketing. The rest of the projects in clarity flow that I'm working on are marketing related. And this is another one where it's like, I don't know about you. I find that there's a bunch of projects that need to get done. Most of them are going to be my projects that I completely do myself, but a couple of them are things that I'm going to hand off to my team, like that feature. She's going to be working on that. This other project is we're migrating our help docs from our subdomain on help scout into our main domain to make them part of our actual marketing site. We can get into that, but that required me first to design and build the section of our marketing site that's going to house our support docs. It's a little bit of design and code work in our statmx CMS. We've got a searchable knowledge base that is now built into our marketing site. I set that up. I designed and shipped that functionality. And now Cat on my team is going to be tasked with actually migrating all the docs, editing them and doing some things. [00:13:07] Speaker B: Can I ask what you're after there in terms of benefit? Is it SEO? Is it. [00:13:12] Speaker A: It's SEO. It's two things. It's, it's SEO and it's, I'm starting a big effort on clarity flow to make the whole marketing site just much more product focused. And we're going to be removing a lot of the contents, unhelpful content, if you will. [00:13:33] Speaker B: Okay. Okay. [00:13:35] Speaker A: Just making, just, just getting the focus so, so much more squarely on. Like this is the product, it's this type of software product that you, that, and here's what you can do with it. And I think that a person who is our help docs, which a lot of them have videos and very detailed, like what you can do with the product and how you can do it with the product, those really should. And they often do serve a dual purpose. They help our existing customers and they help sell new customers and look at what you can do with the product. I think that there are a lot of customers, especially in clarity flow, who, they spend a lot of time to dig around our docs and our videos and learn about like, what can I actually do with this and how does that actually work? And does it actually work the way that I think it should work? And they're digging into that stuff before they even start a trial, or at least on the first day of their trial, they're checking that stuff out. So I think that we, like, one of my marketing realizations now that I'm starting to learn is at least, and it depends on the space of your product, but in my case, it's, um. I think we, we tend to think that like, people come to our SaaS marketing site and they don't need much in order to start a sign up. Like, like sign up for a trial. You just need to sell them on a headline and a benefit and a pain point and like, okay, if they identify with that, like, that'll be enough for them to, because there's nothing to lose. [00:15:12] Speaker B: It's not a big deal. Just create an account or. Yeah, I was gonna ask you what. [00:15:16] Speaker A: I think that, I think that that's wrong in a lot of cases, maybe not all cases, but like, at least because I find that the customer wants to spend a lot of time digging into the details before they are sold on the idea of like, okay, it's worth my time to actually proceed with a trial. [00:15:34] Speaker B: This is the thing I've been looking for. I was gonna ask you if you had a way to describe the way the site is currently compared to what you're describing it as going toward, which is very product focused and pulling these things even from a different domain, pulling it all the way out front to let people do the research that they want. [00:15:55] Speaker A: Yeah, yeah, some of it. A lot of this is not implemented yet, so, you know, it's not going to be a major change visually. The style, the design is all going to be the same, but the organization of the content and the organization of the copy and the focus of each individual page is going to be different. I think overall we're going to have fewer pages. We're going to be killing and redirecting a lot of pages. We are going to, yeah, I mean, that's, I think, yeah, like the, I think the help doc, bringing the help docs into the main site is going to essentially replace the need to have multiple product feature pages. So right now we have a homepage and then that links off to maybe five or six pages dedicated to specific parts of the product. These are product marketing pages. These are not help pages. They're other versions of our homepage. And then in addition to that, we've got a lot of like comparison pages. We've got like reviews pages, we've got all these and then we've got the blog. But the reality is most of that stuff is unnecessary and bloat and pages that nobody ever even wants to click to or link to. But the help pages, the detailed tutorials with videos, people do spend a lot of time on these pages and they find them on Google too. And it's not like, logistically, it's like, why duplicate the work? But it's also like, I don't know, I think people actually want more details than we give them credit for. [00:17:53] Speaker B: Right. The difference between marketing page that talks about a specific feature or category of features and the way we make that pretty, and we don't dive into too many details and we'll write a sentence on how something that you can accomplish with the product, the difference between that and a help doc that digs in into the real, real details. That's a pretty interesting thing to note. Like if, you know, we've talked over the last year. Like what the hell works in marketing? How are people buying what is happening out there? And it feels so chaotic. That's kind of an interesting tidbit to look at and say, all right, if that works better, then that tells us something about how people make their decisions. [00:18:33] Speaker A: Yeah, I think a lot of people make most of their decisions on your homepage and no other pages. But if they do spend time on other pages, they're probably looking for like deep dive details. They don't want like a second homepage or a second like. [00:18:49] Speaker B: Right. Or four homepages for each feature category. [00:18:52] Speaker A: Yeah, and that's kind of what we have now is just like a lot of versions of our home homepage. Another thought that comes to mind here is think about developer tools, developer products. Yes, it's true that developers are a different animal and they consume products and you sell to them a little bit differently than other types of customers and types of products. But I do think that there's something to be learned from developer products. I think that the, the big, like the best, best practice if you are selling a tool to developers is you have to have best in class, just world class documentation. Look at the, I think of like the tailwind CSS site, their docs, as like the gold standard. [00:19:40] Speaker B: Yeah, that's a lot, though. That audience is unique because they think uniquely, they read uniquely, they, their peer circles are different, their decision process. I think that's kind of like the most extreme version of. You must give me a lot of detail. [00:19:55] Speaker A: Yes, but what I'm saying is their docs, their website is so easy to navigate, so easy to search and find the answer to, and the content itself, the text, the images, it's all so easy to consume and easy to, to go down these rabbit holes. That's what developers do. I think that other customers do that. [00:20:22] Speaker B: Too, in their own way. [00:20:23] Speaker A: In their own way. I'm not talking about technic. They're not going to read code tutorials like someone on a talend CSS doc site would. But they do want that because oftentimes they're switching from some other tool. They need a good reason why it's worth the effort to switch from another tool. It has to functionally work a little bit differently and better in some ways. So they're going to dig around in the docs or whatever videos they can find to say like, oh, that has the little function that I've been looking for. Maybe it's worth the switch. [00:20:57] Speaker B: Very cool. I'm excited to see how it works and I'm excited to just the fact that our website is so new that we don't have anything. I'm going to try not to make mistakes that we need to go back on. We're definitely launching with a landing page and not much more, but it's not bad. And the design, I think, is coming out great. And that's really exciting. I have this, maybe, I don't know if I talked about last week, but I've been thinking about it. New competitors are launching. I think I talked about that, a 16 Z article that came out about voice agents. And you look at it and it's like, it's a bit thin in our category and I'm assuming it's going to get filled in. So I just keep an eye out. I follow a few Twitter accounts that talk about AI. I'm subscribed to this newsletter about AI companies. When I tell you every day, it's seven days a week, $115 million raised, $44 million raised between these companies, $280 million. It's just relentless. [00:21:55] Speaker A: I'm on the Ben's Bytes newsletter. Are you on that? [00:21:59] Speaker B: Yes. Yes. That's one of them. Yep. [00:22:01] Speaker A: Yeah. [00:22:02] Speaker B: So it's pretty intense and I'm kind of preparing myself for that level of competition. And recently one launched and what I found myself doing is like, I was going to be disappointed either way. Like, when I, when I, okay, I see it. I read about the competitor. I'm like, all right, that's kind of pretty firmly in our space. And I go to it and I'm expecting like, one of two things. It's either not impressive, in which case my reaction is based on that, it's like, oh, whatever. You know, like nothing to worry about kind of thing, or it's really impressive. And then you get that weird, dark feeling of, oh, shit. So I'm almost, I'm just waiting to be, I'm waiting to come across a competitor that's really impressive and to get that sinking feeling it hasn't happened yet. [00:23:00] Speaker A: What do you think makes it impressive to you? Is it just the overall design look and feel, or do you see some indication of like, oh, they have a lot of customers? [00:23:10] Speaker B: Yeah, okay, that's definitely one of them. But so many things are so new that they just don't have a lot of customers. And so I think that's a good question. The thing that would give me that sinking feeling it has in the past and other contexts and other companies and whatever else is not so much money raised, that's a little bit of a factor. Just because you have to deal with the money. So someone raises $50 million, you kind of roll your eyes, like, now we're going to deal with that. They're going to advertise, they're going to be all over the place. So that's kind of, that can be frustrating in that way. Regardless of how good or bad the product is, you just have to deal with it. That's what we had with Bolt, which was a terrible product. But to raise $900 million, you're like, they're going to be in every deal. Their salespeople are going to be emailing people. You just have to deal with it. So that's one factor, but that's not the most important factor. I think the thing that gives me the biggest sinking feeling is positioning. If they are describing things properly and highlighting their features in the right way, that worries me because then I'm like, oh, they got it right. [00:24:19] Speaker A: It's like, they get it. [00:24:20] Speaker B: Yes, yes. So if it's beautifully designed, that's one thing. If the product is impressive, then the customer list, maybe the money race, but the thing that worries you the most is, oh, shit, they got it right. So that, that is the thing that I'm hanging on to right now, where no one has taken the angle that we are taking on the product itself and the positioning of that product and how technical or not technical, how easy or not easy, the pricing, the set of factors that plot out the exact position to attract that type of customer, no one's there right now. So that's me being happy about that and just waiting for the shoe to drop. Like, uh oh, it's going to happen eventually. So rock made a good point the other day. I expressed this to a few people and rock was like, jordan, that's going to be us. [00:25:12] Speaker A: We're going to, like, people are going to look at you like, oh, shit. [00:25:15] Speaker B: And I'm like, yes, I forgot that that's exactly what we're going for, actually. But for some reason, you kind of view it. [00:25:21] Speaker A: That is a really, really good point. The other thing that I think about for me, probably the reason why I don't spend a lot of time thinking about or looking at the competitors very much is I just assume that there are competitors in anything that I do. But going back to why you chose this market, you chose a market that is huge. Like, it's massive. And so I'm always of the mindset of, like, like, I chose this because it's big enough. It's. It's big enough for me. Plus other competitors to. Yeah, to be fine. [00:26:01] Speaker B: You know, I think I've. You know, I've. [00:26:03] Speaker A: I know it's a different game in the VC games. [00:26:05] Speaker B: Cooling a little bit on, like, yeah, cool story. But you really, really want to be out in front as number one because you just win so much of the market from that. So as long as you're not established, you're not comfortable, and then maybe once you're established, you're still uncomfortable in a different way. But the opportunity to become number one and then get the partnerships and the integrations done and, you know, just that leadership position makes everything so much easier. So that's kind of what I'm after. And I definitely drank the cooling after raising money. You're like, yeah, that's how you do well in this version of the game. So normally, I would say, yes, there are definitely a thousand customers out there. It doesn't matter what competition is out there. And therefore, I can build a business. In this case, I'm a little. This is where raising money changes that mindset. [00:26:53] Speaker A: Yeah. And also, your decision to do an AI product like this, like, so much of the game is being early. Right. And part of that game is, like, being first. Being the first one that people talk about. [00:27:05] Speaker B: Yeah. Or at least the first one in that specific position in the market. There are more established call center AI products that are already series b and beyond, and that's a different thing. So it's just this spot that I want. [00:27:21] Speaker A: A question came up on Twitter. This is also a thing that I was talking on Jessica Malnick's podcast the other day, which that was a show all about AI. [00:27:33] Speaker B: Tweet about that. That sounds cool. [00:27:35] Speaker A: Yeah. I don't know why I was on a panel talking about AI, because I feel like, so far from an expert, but I think it's interesting. I was just talking about. What I think is interesting to observe, at least, is the places where I am heavily using AI day to day and the places where I want to be using AI, but I'm. I don't think it's good enough yet. [00:27:59] Speaker B: Okay. [00:28:01] Speaker A: One of these things I was just describing how we have a project going on where we're migrating a bunch of our help docs out of one system and into another system that involves, I don't know, 75 plus different pages of content in one system, each with multiple fields. So there's a lot of, like, highlight everything here, copy it, paste it into this other CMS system over here. But while you're at it, let's make some edits. And adjustments to the content. So it's like 10% thought analysis work and 90% just data entry, copy and paste. But it's also between multiple browser tabs and multiple tools and multiple page refreshes. And click this button multiple times. And so that's the kind of like, sort of like low level, like tedious, tedious work that like, my first thought is like, okay, someone's gonna need to do all this work. My first thought is like, maybe I'll just hire a va, you know, to take care of that. Like, you know. Cause the only other person on my team who is involved in this is Kat, and she's our customer success person. She's higher level, she, she works with customers, she's strategic, she does some marketing work. So I don't really want to task her with that type of work. I feel like it's below her a little bit. It's not a really good use of her time. I'm not going to be doing that work. So my first thought was, let me hire a VA for a temporary one week project. Give you some instructions, just copy and paste and that'll be that. Then I was like, but it does require a little bit of strategic work that Kat really should do some editing, some rewriting, some reorganizing, and some thought work into this. So I want her involved in the project. I was like, there should be a case for AI here. I should be able to use a tool to do a workflow across multiple pages, multiple tools, multiple page refreshes, copy and paste, all of that. But that tool does not exist yet, or that capability does not exist yet. [00:30:15] Speaker B: You know, kinda. I've seen two products over the last few weeks that you can train based on what you're doing on the screen. So it's basically like, watch what I'm about to do. Click here, pull down tab, select this, hit enter, wait for it to refresh, type something out, copy, paste, go to this screen, change tabs, and it can be trained on your actions on your computer as opposed to something separate outside of your machine, that it goes to a server and it learns something and it gives you feedback. So I think all those things are close. [00:31:00] Speaker A: I think they are close. But here's another trend that I'm noticing all the time now with all these AI tools, they are over promising and under delivering and under delivering, like big time. [00:31:12] Speaker B: It's a bit dangerous in that way. [00:31:14] Speaker A: I mean, I see it every day with developer tools, with, I mean, every single day, like, but also what you're talking about I've seen those as well. Like trained it on this, on clicking around, and I see that, and I'm just skeptical because it's like, that's pretty cool. But I'm just thinking about the amount of time that I'm gonna spend troubleshooting that and like, why didn't it do it correctly and this or that, I just don't think it's there yet. If I think about the developer tools or even chat GPT itself, so many of them are marketing themselves as like chat with your code base. GitHub, copilot or cursor or any of these has access to your entire project. All the files in your code base. It should know if you ask a question, if I'm working in this one controller over here and I ask it a question, it should be able to access my routes file and these other models and these other files in my code base to know, okay, those definitions and those exist over there. Let's help me inform my answer to Brian's question here. No, it does not work. And they advertise it like that. They're like, chat with your code base. It's plugged into your whole code. Like, no, it actually does. Clearly it's not aware of whole parts of my application that exist in these other files, but it did not take those into account. When it gave me an answer, it told me it knew, but it just clearly does not. Chat GPT has the feature where you can attach files to it. You can attach a spreadsheet, attach a PDF, and give it a prompt and say, read these files and then tell me what you find with an answer to this question. So the other day I gave it, I think, like four file attachments. You can put multiple file attachments, I put like four, along with a question, like, reference these four different documents and tell me the answer to this. And just based on their response, clearly it read like the first document and it just ignored the other three. [00:33:21] Speaker B: It's lacquer like everyone else. [00:33:23] Speaker A: Yeah, okay, I get that I'm asking the computer to do a pretty incredible thing, but you are marketing it to me like it's supposed to be able to do this. And no, it's not doing that. [00:33:35] Speaker B: So I think this, I would blame this on the marketing arms race. [00:33:41] Speaker A: Yes, exactly. [00:33:42] Speaker B: You had no choice but to over promise or your product in comparison, looked ridiculously behind. Yeah. And I don't know the name of the adoption curve where it's like the big spike of initial excitement and then the novelty wears off and then there's the trough of sorrow, the wiggles of hope, a little bit more success, and then all of a sudden, like, you know, the promise is fulfilled. So I think we're definitely there where the promises were way, you know, overdoing it, and now there's some novelty wearing off, but I'm pretty sure it's gonna keep climbing, so, oh, there's no doubt. [00:34:20] Speaker A: That, like, it's the curve of advancement is out of control so fast, and that's what makes this whole thing so exciting. Like, think about where we are today and where we're going to be a year from now. Think about where chat GPT was a year ago. It's incredible. Yeah. [00:34:37] Speaker B: And I'm learning more about this. When I heard the term prompt engineering, I really didn't know what to picture my mind around what that was. And as our team learns more about this, I'm basically asking to be educated by the engineering team so that I have a better understanding. I can describe it more properly. I think about how to market it. I could not sound like an idiot on podcasts. And so on this podcast in particular, I'm perfectly fine sounding like an idiot. So I'm going to talk about one area that I learned about this week that gave me a good a better understanding around how to improve the quality of responses when you're trying to accomplish something. So I learned about the difference between system prompts and user prompts. [00:35:27] Speaker A: Right. [00:35:28] Speaker B: And that was very helpful for me to just understand that it's not about the words that you're asking. It's also providing the larger context to the model and saying, this is what you need to know. This is what you need to use as your foundation of information. This is what you need to ignore. This is what I want it to be like. This is my goal. Those are all the system prompts, and then the user prompt is look into my docs and give me this. So those two things combined, what we've been talking about internally, is differentiation and building up value over time. And I didn't really quite understand how prompt engineering fell into that. But our ability to improve over time is one of these things where, okay, maybe you can build a product just like ours in a short amount of time. You can't shortcut the amount of pain and learning involved in how to set up the system prompts to make success in the individual user prompts much more likely. So if someone says, actually, I changed my mind, next Thursday isn't good, can we do the following Thursday instead? So that is a user prompt that goes into the system, the system prompts that we build into our model, then make it much more likely that the system will understand what the following Thursday actually means instead of giving a bad response. [00:36:59] Speaker A: Yeah, I like that. That makes a lot of sense. [00:37:02] Speaker B: I think a lot. When I hear people say it's not good, I always have in the back of my mind that's not that it's user error, but there's a lot that goes into making it successful. So a lot of these larger companies that are hiring prompt engineers, you're like a what? You're gonna hire someone who types a slightly better phrased question like, I don't. I didn't understand it. [00:37:25] Speaker A: Yeah, I was. You know, you're right that I feel like I'm very far behind the curve on the whole idea of prompt engineer. I know what you're talking about with system prompts versus user prompts, or whatever the front end prompts are, but even in my chat GPT account, I put in some baseline. They have a feature in there where you can say, I generally work in Ruby on rails, so when I'm asking about code, just assume that it's a rails project. [00:37:54] Speaker B: Is that what's called memory? [00:37:56] Speaker A: I'm not really sure. [00:37:57] Speaker B: Maybe feature. [00:37:59] Speaker A: But what you're describing is building an AI product. So I'm sure using the API, you can feed it system level prompts. So I was on this podcast the other day, Jessica Malnut's podcast, and I was on with Matt Pritchett, who runs a tool called Quillbee, which is really interesting. I haven't really used it, but I know that they've been launching it over the past year or so, and it's one of these AI writer tools. My hope for it, or at least in general with AI, was to be able to use AI to help me produce more personal content that I might put out on my personal Twitter or my newsletter or my blog, but I've never even come close to feeling comfortable actually publishing something under my personal name that was AI generated. [00:38:56] Speaker B: Okay? [00:38:57] Speaker A: And I've tried multiple times. One time I was on a plane, on a flight, and I had an idea for a blog post. So I just put down maybe like ten bullet points of what I want to include in this thing. Then I fed that to chat GPT. I was like, can you make this a blog post? And here's copy and paste. Like, three of my previous blog posts make it sound like me, and it just did not. [00:39:21] Speaker B: Yeah, not yet. [00:39:22] Speaker A: And that's probably just too basic. That's not like a prompt engineering thing, but the thing that I was hoping for. I don't know if tools like this can do it, but I want a much more limited AI chat GPT or some kind of like a far, far limited thing, like intentionally so, even though I told it like, hey, these are the bullet points that I want to these are the points I want to make. And here's a sample of my writing. I don't want you to come up with any new topics. Don't come up with any new points. Just use my points, just write them so that it's like a readable blog post. It still went out and like added more content to my article. Or it's still extended sentences, like an extra sentence or two with something else that like, no, I just don't, I don't want that. And, and it also still just doesn't sound like me. Yes, these are readable sentences, but they are not sentences that sound like my voice because it's from my personal blog, right. Even though I tried to train it on, on that. So what I want is a much more limited thing where it's like, it literally does not have access to the rest of the Internet. It only has access to my blog, my newsletter, archives, maybe like archives of this podcast transcripts and like that's it. And that's all it knows. I realize it's going to be very dumb. Like it's not going to be able to generate, it's not dumb, it's specific. It's like I don't want it in some ways just for this personal, just for this use case of like write my personal blog. I still want the personal blog to be my ideas, my bullet points. So I don't want it to generate content for me. [00:41:08] Speaker B: Right. Right. Don't get too creative on me. [00:41:10] Speaker A: Don't, just don't be creative, just. But I would like it to take care of the legwork of writing a clean, easy to read blog post. Yeah, like I don't want to do the polish work. I just want to get my ideas out of my head and published the. [00:41:26] Speaker B: Draft and then let me just review it and click one button. [00:41:28] Speaker A: Let me give you like a super, super raw, rough brain dump of bullet points and, and then like publish it. Something that I could be proud of that actually sounds like my voice. And because it's only trained on me, like maybe there's some way to do that. I just haven't had time to like find out how to do it. [00:41:45] Speaker B: I don't think it's far at all I, the, you know, the, what's the term? Oh, just agents. That is the term I hear for this type of thing. And, you know, people can't help themselves. They go Sci-Fi on it. And they'll say, my agents will be talking to your agents and they're going to be creating agents, come up with. [00:42:03] Speaker A: But take over the world. [00:42:05] Speaker B: Yeah. In reality, it's just these very purpose specific set of agents. And that I don't, I really don't think that's far. I think there's a personal assistant type agent that's coming. There's still, you know, silly things that we do that are a giant waste of time, like the ability to just open your mail and just scan it. Oh, yeah, just help me, man. You know, that that's really what we want. [00:42:33] Speaker A: Oh, my God. Do all the shit in my life that I don't want to do. [00:42:36] Speaker B: I had one of those moments yesterday. God darn it. You know, it's like I go my three months without opening my mail. [00:42:41] Speaker A: Me too. [00:42:41] Speaker B: Sorry. Sorry, everyone. And then, and then I bite the bullet and I just open all of it at once. And when I tell you the first piece of mail, the first on. Look at this here, look at this. [00:42:51] Speaker A: This I'm showing my stack that looks like my counter. [00:42:55] Speaker B: Everyone's got a damn stack. Yeah, the first one I open. Maybe I should have known not to open the IR's one. [00:43:02] Speaker A: Oh, yeah. That gives me so much stress. Like, like there's a, there's an envelope that's from the IR's. It's sitting on the counter and I don't touch it for like weeks. Yeah, I just, like, I don't want to deal with it. [00:43:12] Speaker B: The first amount due immediately. 8900. God dammit. And I know it's wrong. I know it's wrong. And I have to go to my account. If they're stressed out for two days and they come back to me and say, we paid that in 2023, don't worry about it. [00:43:26] Speaker A: Anyway, I got a letter from the IR's the other day to state, like, we've received your request that you made eight months ago, but you missed a detail and you have to resubmit something. I'm like, I don't even know what it was eight months ago. What are you talking about? They sent me a mail, a thing in the mail to tell me this. I'm like, what? [00:43:48] Speaker B: Okay, well, we digress. Here's what I want to talk about more AI. But first I got to give a shout out to a new person. I follow on Twitter my man, Cody schneider. This dude is on God level amounts of Zinn. I don't know what is happening in this person's brain. He is an idea machine. Follow this man. There's so many good ideas. I keep. I keep pasting them over into our slack. Eventually, everyone's like, what is up with this guy? There's so many freaking ideas. So Cody Schneider runs a product called Swell AI. And I think this. [00:44:31] Speaker A: I think I just heard him on a podcast somewhere. [00:44:33] Speaker B: Yeah, he's been on a few podcasts, too. This thing is where I think small software companies like us, like, we have to go up this learning curve, because there is now a way to take one or two people's marketing efforts and produce an amount of work that ten people would be required for. And that feels like an absurd thing to ignore. Absurd. If we limit what one or two people can do on marketing and not use these, like, amplification products. So what swell AI does, for example, is we could take this call, this podcast. We could take the video of it, because we're on video, and we plug this thing into swell AI, and it will amplify, it will create 40 different video clips with the subtitles ready and formatted properly for TikTok, Instagram reels, whatever else. It'll schedule 50 tweets, and you can go in and accept or decline whichever ones you want or don't want, or schedule them. And this is what I told Sam, who is basically my go to market. Everything on our side. We have to learn this. We have to be able, even if the quality is not as high, it will not be, forget if, it will not be as high of quality as if we did everything ourselves, the old school way. But the quantity having a quality all its own mindset, I think, is required. So I'm looking at my boy Cody, who comes up with these amazing ideas, looking at his software product, and I'm looking at David from Jenny AI, that blog post that I'm kind of looking at as, like, this north star. It's the combination. That's what we're gonna do, I think. [00:46:30] Speaker A: I like where you're headed with this. And I've wanted to do this for so long, to have a pipeline of content that is just like, I create one thing and then send it out as tweets and LinkedIn posts and newsletters, whether it's people, whether it's AI. But I feel like the thing that, the key thing to make that work is you have to have really good source content for the first piece, right? Like, if we actually cared about this podcast and made this like a business enterprise like, content thing, I feel like we would have, like, a lot of source. We could take each week's episode and we're not doing this, but that could turn into halfway decent streams of tweets and LinkedIn posts and newsletters based on what we actually talked about here on the podcast. So I feel like if a company's gonna gonna use something like this, like a swell AI, like, you still need to have probably the founder, I think, or someone high level, someone could like doing a high quality podcast or YouTube channel as like the primary source. And then everything else is what I'm. [00:47:49] Speaker B: Repurposed, what I'm hoping. I agree with you 100%. What I'm hoping is that, like, chicken and egg issue is solved with the motivation to create the initial good content being there. Because look, if I were to start a podcast for the new product, and then went out and did all the work required for a podcast, and then did episodes, and then all we did with it was publish a podcast episode, tweet about it once or twice, and send out an email about it once or twice, it's almost like not enough motivation. I just don't believe that's going to make that much of an impact. So it's not motivating to do the initial work. I'm hoping that the amplification opportunities provide more motivation to create better source. [00:48:40] Speaker A: I mean, I also sort of agree with that now that, like, even amplifying it and even just content in general is not going to do a whole lot in terms of, like, gaining new exposure, you know? And I think that a lot of these tools like this, like, because I have used these for years, but even before AI, like, there's, there's always been the idea of like, scheduling social media posts, like, tons of tools that have been around for decades doing that. Right. For the most part, those scheduled, repurposed social posts do not do well. The ones that are actually personally, from a person with a unique point of view, something controversial or something super interesting or something that gets a reaction. That's what actually goes, whatever, viral, you know, the repurposed, the formulaic headlines, the clickbaity stuff, I think that that is getting less and less effective these days. [00:49:44] Speaker B: Yeah, I agree. I'm not sure if, I'm not sure if there's an issue there between personal and business, though, because I don't want to hear an individual person's like, regurgitated, rescheduled, whatever. But a business that's offering me something that I'm going to be interested in. I. [00:49:58] Speaker A: But I feel like a business even more so. Like, we tune them out even more than people. [00:50:03] Speaker B: But, but I think what works there is just ubiquity. [00:50:08] Speaker A: Just being. Branding. [00:50:10] Speaker B: Yes. It's almost, it's almost like a b to C product that you need to see ten times before you're like, all right, I guess I'll go check this thing out. [00:50:21] Speaker A: But then the question is like, okay, I buy that for consumer products. I don't, I don't know that I buy that for b two b products. [00:50:31] Speaker B: I don't know either, but we're gonna find out. Cause that's, that's how I, that's my approach. That's how I think this product is best distributed. Yep. Yep. A slightly indirect approach. Right. People on Instagram reels and TikTok are not there for business, but there are a lot of ads for business. It's a normal expected thing. I think that environment plus ubiquity is what will give us an opportunity to get the actual offer in front of people. [00:51:08] Speaker A: You know, this sort of gets back to where I'm headed marketing wise, with, with clarity, flow and everything. I just, I just think that b two b, especially b two b SaaS is so much, so simpler is, is a misleading word. I'm not saying simpler like it's easy. I'm actually saying simpler like it's harder to market b two B SaaS. But, but the way that, the way that customers buy b two B SaaS is, it's a much simpler approach. They, they tune out the bullshit. So we can't market SaaS. We can't market b two B SaaS like their consumer products. And we also can't market them like they are content products. And I think that a lot of SEO advice and agencies and consultants will convince you you got to do all the SEO content things, just like a info product site or a content site or a news site would. But b two b people are searching for product software and then they might talk to their business colleagues and about, like, which software do you use for this or that? And then they recommend. And other than that, like, it's okay. Maybe there's like a, there's also like the, like the sales play of like, going into companies and like doing, doing the sales cycle. But the idea of, to me, like the idea of like winning through Instagram ads or TikTok or branding or viral, it's, it's hard to buy it. [00:52:45] Speaker B: You know, I hear you. But I think that is, it goes along with our situation. Awareness is super important. Right. Because the search volume isn't there yet. People aren't actively looking to solve this pain. They're just starting to hear that it's. That it's a pain that's solvable in this way and so getting out directly in front of them, but using those channels almost to point out the pain, to identify with the fact that, oh, yeah, other people have this problem also. Yeah, that's generally what I'm thinking. So it is user generated, content based advertising. A whole bunch of words mashed into one. But you know what I'm saying. The ad doesn't feel like an ad. The ad feels like someone just like you who had this problem and is talking about it. [00:53:38] Speaker A: You know, this sort of ties into something I've been tuning into a lot of lately. Do you follow Matt Panisi? He runs a site called Money Lab. He's had it for a while. Just really interesting. Like, I don't know him, but personally, but a very solid content creator, really solid SEO practitioner. He runs a site called Swim University, which is a long time content site for swimming pool maintenance or whatever. So they've won the SEO game in that space through content and affiliate products and digital products and things like that. He has this other side business called Money Lab where he talks about how he runs online businesses. He's been doing it for a while, so I really like his stuff. And he's got a podcast where it's just him talking solo, just brain dumping these ideas. He's been sort of on the forefront of his whole business is based on SEO, and it has been for over ten years. I really trust his opinions on SEO, and he's been very bleak about where it's headed now with the introduction of AI and how, like, just look at Google. Like the front page. Like, the top half of Google is not just sponsored spots anymore, it's now it's AI from Google. [00:55:08] Speaker B: The whole thing is so weird. [00:55:09] Speaker A: So it's like, what happens in a world where Google as a thing goes away? Like, the company's not going anywhere. But I'm saying, like, Google, like SEO as a channel, think about how many of our businesses depend on that being a thing. I don't think this is happening this year or even next year, but sometime in the near future, there's going to be a future where when people have a question or when people are searching for something, they're not going to Google, they're just going to type into their iPhone or whatever device or tool they're using, and they're going to ask the question and some sort of AI service is going to respond. Next week, Apple is about to announce their integration with OpenAI. Like, it's gonna be part of Siri. So, like, there's going to be a world in the very near future where, like, people are not googling for things anymore. They're just asking and getting responses from whatever they're holding in their hand, you know, and what, like, my question is, like, what is that gonna mean for us businesses that depend so heavily on Google for visibility and exposure and access to customers? You know. [00:56:27] Speaker B: I don't know. I don't know. There might be a lot of opportunity in that transition. [00:56:35] Speaker A: Yeah, I think there's something that replaces it. I don't know what that's going to be like. I think the thing that most people tend to start to think about is like, okay, well, if we have to start relying less on Google, maybe we need to start to build up our personal brand on YouTube because people trust people. A more human approach. I happen to think there's going to be more of a referral based approach. And that's why I'm investing heavily in customer success, because the best customers are referred from other customers and, you know, lower volume but higher quality subscribers. That's not the answer to replace Google as a channel. But, but I'm just saying, like, at some point, all these businesses are going to have to. It's going to be a new world and it's coming sooner than we think, I think. [00:57:31] Speaker B: Yeah. The only comfort is that most people are much slower to adopt than we are. So that that will tail off over a longer, relatively long period of time. [00:57:40] Speaker A: Yes, but, but just look at how unused the current behavior from the biggest companies, Apple, Google, Microsoft, Amazon, they are doing very unusual things. Google is completely redesigning their Google search results page. That's the core of their whole business, and they're completely messing with that with AI. Apple is about to announce a huge partnership with OpenAI. Probably this stuff is gonna go mainstream a lot quicker than we think it is. Last night I was watching the NBA Finals. There was multiple tv commercials from Google with Jay Z endorsing it, promoting Google Gemini, dedicated tv commercials on the NBA Finals showing consumers, hey, AI is here. You could use it now, you know. [00:58:39] Speaker B: Oh, boy. I don't know. One year at a time. [00:58:43] Speaker A: Yeah, fun times. [00:58:47] Speaker B: That's it. That's it. I told my team that I almost completely screwed up the difference between system prompts and user prompts so they are ready for this episode to be published. They want to hear if I got it right. [00:59:00] Speaker A: I want to hear that right. [00:59:01] Speaker B: I told them I wasn't sure when I started talking, but the pressure helped me remember. [00:59:05] Speaker A: I like it. I like it, man. [00:59:08] Speaker B: Cool. All right, folks, thanks for listening. [00:59:09] Speaker A: All right, later.

Other Episodes