Episode 663 | 5 Insights SaaS Founders Should Know About A.I. (Ignore at Your Peril)

Rob Walling:
Welcome back. To start off, for the rest of us, this is the show where we focus on bootstrapping and mostly bootstrapping SaaS companies. Today, I sit down with my TinySeed co-founder, Einar Vollset, and we talk through five insights SaaS founders should know about AI, about OpenAI, ChatGPT, LOMs. All the things you’re probably kind of tired of hearing about. But here’s the different tact that we took that I’m not hearing other podcasts and other pundits make on this topic. I tried to put a mental framework around AI that SaaS founders would understand. It’s how I would be thinking about this if I were a startup founder, and frankly, it’s how I’m advising my almost 150 investments to be thinking about this. So I kick it off by going through four categories of AI. So we had some taxonomy to think about and I talk about generative categorization, summarization, and predictive.
And then we walk through five things that I think you as a founder should be thinking about. If you’re putting your head in the sand and thinking AI’s not going to change everything, that’s a problem because it’s going to change quite a few things. But before we dive into that, tickets for MicroConf Europe this October are available for sale microconf.com/europe. MicroConf Europe is in Lisbon October 1st through the third speakers include Michelle Hansen of Geocodio, Sherry Walling of ZenFounder, myself, Stephen Innala-Craven of Stridist, and several others yet to be announced. It’s going to be an amazing event. Tickets are already flying off the shelves, as they say, microconf.com/europe if you’re interested. And with that, let’s dive into my conversation with interval set on five insights SaaS founders should know about AI. Einar, thanks for coming back on the show.

Einar Vollset:
Glad you have me.

Rob Walling:
It is awesome. We’re going to be digging into AI today, obviously talking about things that SaaS founders, whether bootstrapped or not really should be thinking about. I want to kick it off by maybe giving folks an idea of how you’ve been digging into AI, how you’re thinking about it and how much you’ve been screwing around with it. In essence, you’ve also, and you’ve been doing internal presentations to TinySeed companies. Of all the people I know you’re one of the people who has given it the most thoughts, so to speak.

Einar Vollset:
Well, thank you. Yeah, I mean, I started looking at it I guess in August or September last year, and it wasn’t really on my radar all that much. I have to be perfectly honest, we’ve invested in a couple of AI companies or companies that used AI. It was a key part of it back in I guess 21. And I nearly did my PhD on AI, which actually looking back at it, I’m glad that I didn’t because it was, looking back, it was a terrible time to have done your PhD on AI, but it sort of started, came definitely to the forefront in the fall. And particularly with ChatGPT, it became very evident to me that this was probably one of the bigger shifts in terms of a technical capability that would impact startups. I think at the very least the biggest comparative change that I’ve seen was probably when the iPhone was launched, and I think it’s definitely a bigger shift than that and potentially as big as the internet becoming a commercial commercially viable thing to do.
So I started really digging into it. I thought AI’s an interesting thing. People have weird reactions to it and there are the doomsters out there who are like… We should just call an airstrikes whenever the loss function falls too much. But equally, I think a little bit more of the considered concern meant, well, if you can just ask the AI to do everything, why do you even need SaaS companies anymore? Why do you need needle and stuff? So I started digging into it quite deeply, probably November, December timeframe. And it became evident to me that this was something that we needed to both for the portfolio for TinySeed, but also for the work that I do with Discretion Capital. We needed to lean into it. And probably most of my experimentation today and the systems that we’ve been building has been augmenting our systems for discretion capital in terms of using AI to understand the SaaS, specifically the B2B SaaS market worldwide.
And that has allowed us to basically have a step change. We’re basically gone from tracking the number of SaaS companies that we’re tracking and what they’re doing, and all this stuff has probably gone from 10% of the market to nearly a 100%. So it’s an order of magnitude change and it’s been key to me to try to understand, obviously that’s a huge benefit to us, but also I wanted to understand how it would impact our portfolio at TinySeed. Because we’re invested in so many companies, I felt like there’s probably at least a third, if not more, which there could be a severe competitive impact on the stuff. So I started digging into it. Most of my focus has not been so much on the visual models like the stable diffusion and dollies and stuff, but mostly on the large language models like ChatGPT or Flan-T5 coming out of Google and just trying to understand like, “Okay, how does this impact, how can we do this?”
There’s an odd dynamic in the AI world, I feel like, because it’s on one hand you have people who think it’ll be just snap your fingers and it’ll take over the world. And the other hand you have people who are like, “Oh, this is just auto complete, it’s bull- like why even bother with it?” The most standard thing I hear is like, “Ah, this is Web 3.0. Once again, it’s like NFT craze.” And I’m like, “Brother, it’s not NFT craze. Whatever it is, it’s not that.” So yeah, that’s been my angle. Like I said, my academic background is very much on the systems applied side, and so trying to understand, reading a lot of papers and trying to understand what can it do, what can it not do, what are best practices? And like you say, also trying to get the TinySeed portfolio companies up to speed as fast as possible in terms of understanding what they can expect.

Rob Walling:
And there’s a certain amount of just learning what it can do and not do as you’re saying. And then in three months, six months, 12 months, trying to get ahead around what might it be able to do at that point. Then taking that and viewing it through the lens of SaaS companies, whether funded or not. I mean that’s what we’re trying to do internally. It’s for folks who don’t know, you mentioned discretion capital, that is the sell side M&A firm where you help SaaS companies between one and 20 million ARR exit. You’re in processes. So a bunch of SaaS companies, tens of thousands, literally private equity companies, all this stuff that you do with that. So they know why both tiny seed and discretion capital. And frankly MicroConf and this podcast has an interest in this topic because I keep saying ignore AI at your peril.
I’ve heard the same thing about is this the next web three? Is this the next whatever other technology had a big hype cycle and then flamed out. Yes, there’s a big hype cycle, but I think this is much more no code where no code had a hype cycle and people are like, “I don’t know, it doesn’t really just code.” It’s always just code, you have these [inaudible 00:07:06] but no code is changing things at a grassroots level. It’s changing things for a lot of people. AI is going to do that times a hundred. I think you and I are both in agreement on that.

Einar Vollset:
A 100%. I mean, to give you an idea, it’s rare. The whole NFT thing like that, in my normy friends in Europe, that’s usually my standard measure of how do people interact with technology. And my normy friends in Europe, they never even heard of NFTs. It was like a, “What? Who cares about this? Ran Ethereum, who gives a shit? No code, I don’t care.” But ChatGPT, it was immediately, I was talking to my brother once and he is like, “Oh yeah, crap.” Normally it would take us three, four weeks to write this report that we charge a bunch of money for. And actually, yeah, it can be done in two days now real people are seeing real impact. And I think if you have technical founders who are sticking their head in the sand about the capability just become available, I think it has a massive, massive mistake, a huge mistake.

Rob Walling:
My brother runs a construction firm in the Bay Area, and similarly, he and I talked about it for 20 minutes and I started showing him, I was like, “It’s not like construction is the most high-tech industry.” But he was telling me about briefs he has to write or things he has to summarize or things he has to consume or you know what I mean? It’s just the moment we got in a text, I was like, “Oh, well let’s just… Here, give me that in a PDF so I can copy paste it right in the ChatGPT and just ask it some questions.” And it was a 100%, right? “No.” “But will it be a 100%, right? In the next three, six, 12 whatever, it’s going to get there so.”

Einar Vollset:
It’s going to get there. And I think a lot of the time, technical founders, they underestimate how much of the world economy consists of people who take bits of text and produce bits of text and give it to other people, and those people produce other bits of text depending on the inputs of that bits of text. That’s a lot of the economy in the world.

Rob Walling:
Yeah, it’s crazy. So, hey, when you said you’re Normy friends, so I use the term muggles. Oh, my muggle friends, you know that? It’s from Harry Potter when in the Wizarding school, right? Oh, the Muggles are like the normies, the who don’t have magic. Anyways, I want to kick it off. I put together an outline of my thoughts. The reason I wanted to have you on is I could totally have done a solo episode around this, but I want someone who knows more about it than I am to say, “I disagree. Oh, I do agree. Oh, I agree plus, plus.”

Einar Vollset:
I’m happy to disagree, bro. Always.

Rob Walling:
Oh, I know. That’s to derail. The podcast is your goal. All right. So I have been trying to get my head around the categories of AI that this is not capture everything in the world, but as I’ve seen uses of chat, GPT and Dolly and the OpenAI API, I think that in my head there are four categories, is the only word I can come up with. But the first is generative. This is where you type in something and it spits something out. So that’s like create an outline for a YouTube video titled How To Invest in Bootstrap Startups.
And then it just does an outline. It creates it based on its predictive stuff. Or you can type into Dolly, you can type in create a picture of Einar Vollset with a San Francisco Giants hat on. “Ooh, no, who’s your with a White Sox?”

Einar Vollset:
The Dodgers?

Rob Walling:
That would be great. The Dodgers, “Oh yeah, no. Ooh. All right.” Note to self, putting it on my trailer board, create a picture of Einar Vollset with a Dodgers hat on and waving a big foam finger that says, “I love Tommy Lasorda.” Am I in the wrong decade? Is that us?

Einar Vollset:
I don’t even know who that is. That sounds good to be.

Rob Walling:
Ooh, it’s a deep cut for people who watched baseball in the 80s, they totally know what I’m talking about. But anyways, so that’s what I’m calling generative AI generating things. Second one is categorization. If I feed it 1000 URLs, can it tell me which of these is an e-commerce website versus an agency versus a SaaS app? And then the third I have is summarization, which is a little bit, you could call this generative, but I’m breaking it out because I’m having a ChatGPT and other tools summarize a YouTube video and try to turn it into a tweet thread, for example. So there’s a summary and there’s a bit of generation. And then the fourth one I have is predictive. And this is one where internally I think we’ve talked or brainstormed, just like, “Well, could you input all the inputs of all the successful mastermind that we’ve matched in Microconf?”
And then when the new batch comes in, you put those inputs in and have it try to predict what we should match. Because that matching process is extremely manual right now. And I don’t know that we have enough data and we’ve matched a thousand founders. If we had a million, I would say we have enough data, but I just don’t know there’s enough that it could do it. So those are the four things, generative categorization, summarization, and predictive. I’ve gone to Google and tried to type in what are the types of AI, what are the categories of AI to try to get someone much smarter than me has thought about it in this way, but I don’t seem to be able to find it. Do you have thoughts? Do you feel like I’m on base there? Any other types that I’m missing?

Einar Vollset:
No, I think that’s reasonably fair. I probably think of it slightly differently. I think in general there is this notion, I mean in general, everything is generative. Fundamentally, a lot of the text models are just like, “They’re extremely flexible APIs, basically it’s just text in, text out.” So that’s true. I mean, I think as a broad class as the way I think about, there are classification, which is sort of similar to predictions where it’s like, “Here’s something, which bucket does this fall into?” It could be anything. It could be like, “Is this URL or SaaS app like we do for discretion capital? Or it’s like you say, given this background, does this person fall into this or that mastermind fit?” I think that’s true, and I think summarization is probably also one of the key ways to basically get value very quickly. And one of the key things there is, it’s remarkable to me, you were saying when you were talking to your brother, it’s like it’s remarkable to me how often you can just put dirty data in there and it’ll just clean it up and figure it out.
You could take raw HTL, messed up JavaScript from a website, dump it in there and say, “Hey, give me, what does this business do? Give me the summarize of what this page is.” Even though if you were to do that with pre LMs, then that would be a pain in the to do. And I think for me, that’s one of the key mind shifts that this technology does is basically programming in general is extremely precise. Math is very precise, programming is very precise. I mean, it’s like if you want to interact with text, you better clean up your text. It better be in a specific format. Ideally it’s in the table of some kind. You normalize the data, you deal with it. And it’s much easier as a programmer to deal with exact numbers and things than it is to deal with different various corpus of text.
And I think that’s probably one of the main things to keep in mind when you’re thinking about use cases is that it effectively gives you new fuzzy tools. So it gives you a way to grasp in a fuzzy imprecise way, grab text and text formats and get value from that in a way that is almost impossible to do in a generic way before. Because before you had to handle every side case and every possible whatever, versus with these tools, you can just roughly grab some text and indicate the sort of thing that you want to be able to do and it’ll refine it and like you say, summarize it and put it into a more valuable format where you can interact with it. And in fact, one of the most precise formats or one of the most interesting things is these things around embedding.
So I think that’s the next step that people get to. They start playing with it and they do the thing and then they sort of, “Go. Okay, how do I represent many of this thing? And being able to take extremely diverse textual, maybe dirty text sources and turn it into a very concise textual representation of something that contained information or even better an embedding that you can embed in a database and do things like vector search on.” I think that’s one of the key things to really understand when you’re trying to think about the use cases and how it applies to a SaaS.

Rob Walling:
Yeah, and that’s why it’s helpful for me to have these categories of what it can do, even if they’re all generative, you’re right, they’re all just generative. But as I break it down-

Einar Vollset:
Breaking down the next step.

Rob Walling:
But I’m thinking as a founder, let’s say I ran an email service provider today, or I ran a CRM, I would want to say, when do I ask my users to generate text or to generate images? Well, obviously when they’re writing an email, what are they getting that email from? Are they just trying to summarize something else? So should I build a summarization engine into my ESP, right? Or in my CRM, when I type in the URL of my contact, the sales lead shouldn’t AI maybe try to categorize that because that’s my second category, categorize it and say, “Hey, is this an e-commerce company?”
Or even pull in a summary or write a summary of this company does the X, Y, Z in three sentences such that the sales rep doesn’t have to do that. So having those four categories, and again, I’m not saying if we listen to this in a year, I bet we’ll be like, “Oh yeah, those categories were off. But I think they’re close.” And I think that’s a model of, if I was running a SaaS app today, I would apply each of those four generative categorizations, summarization and predictive to say, where in the app would my users benefit from it doing one of those? Or where do I ask them to do it today that AI can at least give them a start?

Einar Vollset:
And I agree with that, and I think it sort of relates to some of the other ways that I think about it and is what I tell TinySeed founders too is, it’s important given this new capability to not just… It’s completely the wrong approach to think that this is just NFTs and stick your head in the sand that is fundamentally wrong. I’ll take an argument with anybody who thinks that, but even if you think it’s like this is something that is great and you want to utilize, I think it’s such a big change in capability that it’s always behooves you to take a step back and look at the businesses that you have and the customers that you have and the use cases and the problems that you’re solving for your customers, and take a big sort of step back and get up to the 30,000 feet level and think about what are the sort of jobs to be done in that generically for my customers and what is now possible given that this capability exists that didn’t exist a year or two ago, what can I do differently?
I think some people will end up not getting the most of this or be left by the wayside because they’re a little bit too close to their existing solution. They’re like, “Oh, I’ll just add a chatbot to my thing, or we can do this.” And it becomes very hard to build any defensive mode around that because it’s like anyone can add a chat chat. There’s got to be, at one point I was looking at product hunt and it was every single thing except two things on that day’s product hunt was chat to my PDF docs.

Rob Walling:
It’s too obvious.

Einar Vollset:
It’s like 12 different ones. It’s like, “Yeah, okay.”

Rob Walling:
Exactly.

Einar Vollset:
Well, this is the tutorial case for almost all large language models like, “Come on. This is not a product. This is an auxiliary thing.” And it’s important that you take a step back and say, “Okay, what problems can I now? Maybe I can solve the same problems or bigger problems for my customers than what I was able to do before.”

Rob Walling:
Yeah, that’s our first point really. I have five points in an outline today. We may do more depending on other thoughts you have, but really the first one is to take those four categories that we’ve just talked about and ask yourself where each of them could be applied to your SaaS to help your customers to just make it better is to mental model. And as you said, the obvious ideas of putting a chatbot is not going to be enough.
My second thought or the second point that I’ve realized ties right into what you just said with the chat PDF and how eight out of 10 on product hunt were that unless you build something novel that is non-obvious and relatively difficult to build, AI is not a differentiator if all you’re doing is engineering a prompt and you can build it in a weekend, even though it can do something totally cool, anyone else can do that next weekend. That’s a big mistake. Everyone can use it. The obvious idea is, again, summarizing X, Y, Z or building a chat for your PDF, there are going to be hundreds of those, so you have to go further and think about moats are still moats and five hours of code is not a moat.

Einar Vollset:
Yeah, I definitely think that’s true. I think basically the key thing to think about in terms of how I think about founders and their SaaS businesses is like you say, I think it’s important not just to think, “Hey, don’t just stick your head into that.” If it’s two hours worth of work and it adds a really cool capability to your app, you should definitely do that. It’s a mistake, not two, but you can’t think that that’s a competitive advantage, an additional mode than what you currently have.
What you really want to be doing I think, is to say, “Okay, I have my mode. How can I further add to it with AI?” And I think if you don’t do that, if some obvious stuff that people eventually will come around to realizing, “Oh, we should add this capability that the LOM gives us to the problem domain that you’re working in. If you don’t do that, then someone else will and someone may come out as left field and basically because you refuse to go with the times and add this capability to your product, then that might be a competitive advantage for someone else because they choose to do it and we refuse to do so.”
There’s going to be some failure modes there where people are sticking their head in the sand about this and being like, “Oh, it doesn’t matter, whatever.” I think a lot of people, not a lot of people, but some people are going to get out competed by this low-hanging fruit. But the flip side is like you say, the low hanging fruit isn’t in itself emote,

Rob Walling:
It’s more an accelerant. I think of it, I mean there’s a ton of examples we could use, but remember when rails came out for Ruby and somebody… I don’t remember if it was DKH or someone on the team there built Twitter, a small version of Twitter MVP version in 20 minutes on a video, and people were like, “Oh my gosh, now I can build Twitter in 20 minutes.” And it’s like, “Yeah, that’s cool. That’s no longer a differentiator, like the code to build that.” And so if you were still writing Ruby with no rails and building web apps, suddenly you were way slower than everyone else. And in fact, if you were using… I mean, there’s a reason Laravel came out with PHP, Django came out with Python because those languages became much, much slower than Ruby at building web apps. And so again, it’s an accelerant. It feels like, “Oh my gosh, I’m ahead of everyone, but it’s like everyone else can use this too..”

Einar Vollset:
Yeah, it’s true. And the way to think about it, I used the iPhone example early on, this is almost like, which I know is strange. It’s almost like you can, if you have a web app, you can flip lick fingers and all of a sudden you have a native mobile app that works and that’s what it’s like. And if you think about it that it’s like, “Well, it’s so easy just to add a mobile app.” You just click your fingers and you have it, so it’s not a mote, but if you don’t do that, someone else will. And then that will be a competitive advantage for them.

Rob Walling:
Becomes table stakes in this space, right?

Einar Vollset:
Yeah.

Rob Walling:
All right, so that was the second point. The third thing, I’m curious if you agree with me on this, but I wrote this out, the big AI ideas, trying to build your own models, your own LOMs and the massive horizontal plays, building a search engine with AI, it’s like these are big, these are going to be billion and billion dollar companies, I think those are already done. They’re going to be won by OpenAI, by Google, by Microsoft, by Facebook, IBM, whoever else gets into it. These players are so massive and so well funded that if you’re a mostly bootstrap company like the Saas WISI, it’s just too big. What do you think about that?

Einar Vollset:
I agree with that. So for example, one of the things that I have running in just in my local MacBook is I’ve taken Google Takeouts, exported all my email from all my email accounts and basically created a chatbot that allows me to talk to about my email. So I can ask, “Okay, when did I meet so-and-so? Or who are the people that I talked to after I went to this conference?” That sort of thing. It’s very, very cool. But do I think it’s a standalone business? Do you think it’s a good business for a bootstrapper to start? No, because it’s such an obvious thing for the big email providers just to add out of the box. I’m almost a little shocked that Google hasn’t already added away for you to interact with Gmail that is like that. And so that’s effectively, I think the big players will take the low-hanging fruits and accelerate on their way and add that compare, add to the existing competitive mode.
So I wouldn’t want to start an AI email startup type client. That doesn’t make any sense, at least not for a bootstrap type business. Now, if you add Neumann and you raised a gazillion dollars, sure, go for it. But most people aren’t like that. They don’t necessarily do that. So yeah, I 100% agree with that, I still think the same is sort of true for search. Now some of these big hairy ideas that are like, once you realize understand what it does and understand the technology, then it’s an obvious idea. If it’s an obvious idea and a huge competitive market, but large incumbents what those guys are going to take that they’re going to own that piece.
And so building your startups around that and thinking like, “Oh, I’m smarter. I’m more nimble, I’m doing whatever.” I’m like, “You know what? connecting Gmails data with OpenAIs APIs and calling that your startup, that’s not competitive mode.” That’s not going to work for you. You have to go after like, “What is your existing mode, what is something new, something completely new that doesn’t require people to change that behavior in order to be for you to be successful.”

Rob Walling:
Or like we talked about above, you have existing moats probably within your company today. And using AI to make your product better is a way to extend those. And accelerate 0.4 that I’d like to communicate is in addition to thinking about how AI can be integrated into your product as features, which is 0.1, I think most SaaS founders, and frankly most entrepreneurs should probably be using AI internal to their company. Whether you’re doing content marketing or you’re trying to repurpose a YouTube video to Twitter or a YouTube video to a blog post, whether you say, draft a cold email based on my homepage to this type of buyer and have it give you a 101 oh, no, that’s not funny enough. I mean, you can go back and forth with it to help outline YouTube videos. I will admit some of the YouTube videos that I’m putting out.
I go to ChatGPT and say, “Outline a video 12 best business books to read this year, blah, blah, blah.” But then I’ll say, “Nope, not good enough. Regenerate. Regenerate. It’s not even 50% for me, it’s probably 25% I use or a third of what it gives me, but it helps get me outside of my own box.” So we’re using it for content generation, and I feel like small software companies especially really took advantage of when virtual assistances were $5 an hour in 2008. That was a game changer for my little company at the time. And I think that if you’re not using AI in your own internal workflows, again, this is not building it as a feature in your app, but helping it make you faster. I think it’s something that you should be thinking about. What do you think?

Einar Vollset:
Yeah, a 100% agree with that. I mean, I think the obvious case where this applies immediately is code. I don’t write code anymore without AI helping me out, and I probably do only about 20% or 30% of the typing. It just moves faster. I don’t bother if I have a new Python library that I need to figure out how to use, I don’t bother trying to ask Stack Overflow anymore. I just am like, “I don’t read the docs.” My standard thing is, “Hey, I want to use this library in Python. Can you write me up some codes to do it?” For example, I wanted some visualization code. I’ve never been good at a visualization code. And I was like, “Okay, because there all kind of complex. They’re a pain in us to deal with visualization stuff.” So I was just asking ChatGPT, “Hey, can you visualize these 2000 embeddings for me?”
Or whatever it was, and it just did. It was like, “yeah, just cut and paste this stuff in here, make sure you PIP install this library.” And boom, it was doing the visualization, the three-dimensional visualization that I wanted to do without me having to really ever look outside ChatGPT to understand how the library works. And I think that applies. I also think it’s like if you are in a profession or a party or thing is where you’re generating text, for example, reports and things like that, internally, you’re in a massive disadvantage if you’re not using that. Anything that you’re outsourcing to your VA or you have some medium level employee write text for that stuff, you should probably be looking at can, how can I augment this capability with it like an LOM?

Rob Walling:
I have a caveat to this one and I’m curious about it around with you just a little bit. It relates to this one plus the first point about building features into SaaS, but I feel like not every SaaS app itself needs or can use AI inside of it as a feature, but I do think every SaaS company or every SaaS founder could be using it internally pretty much, and everyone internally has something they can use ChatGPT or AI for. But I was trying to think of examples, and I mean there isn’t a great use case for it. What about Ruben’s company sign? Well, right, it’s electronic signature. How could AI be integrated in that? Because I can’t think of a great example off the top of my head.

Einar Vollset:
Sure, I can.

Rob Walling:
Okay.

Einar Vollset:
Hi, Ruben. Basically you have a company, it signs a bunch of documents, a bunch of contracts. How does the company keep track of its obligations that it’s signed? How does it know that?

Rob Walling:
You mean what’s in the documents? Like knowing what it’s-

Einar Vollset:
Yeah.

Rob Walling:
That’s interesting. So it becomes a knowledge base.

Einar Vollset:
All the contracts that are given, I mean contract management is driven can turn into a contract management. You can add a contract management capability to sign. Well, it could be like, “Yeah, part of the thing, you signed this contract and it says when you’re going to get paid, summarizes your obligations to the client. Did you meet this? Yes or no? It understands every commitment you’ve made and can summarize it for you.” So, yeah no-

Rob Walling:
I like that. This is why-

Einar Vollset:
That’s a good example. There might be, but that particular case-

Rob Walling:
I like that one.

Einar Vollset:
… definitely I can see.

Rob Walling:
How about let’s do one more and then we’ll move on. I think in the world there is a percentage that won’t use, but it’s probably pretty small.

Einar Vollset:
I’m sure.

Rob Walling:
Yeah. But what about SavvyCal, right? Derek Reimer startup scheduling link competes with Calendly like-

Einar Vollset:
Absolutely. I want to be able to talk about my SavvyCal account links like, “Okay. Hey, summarize, what did I do with my… How many times have I talked to this person before? When was the last time I talked to these people?” It’s unstructured, semi-structured data about what I did with my days that can be generated in text? Absolutely. You can use that built in. And actually it ties into one of the things I haven’t done, like I mentioned, my email is a good example of how I’m using it to talk about, basically talk to my inbox. One of the things that I’m not doing that’s kind of the low hanging fruit there too, is I want to be able to integrate it into my calendar and just say, so I emailed this person that and that date, and also by the way, I chatted to them, I had a Zoom call with them on such and such a date. That should be part of the information that summarizes into what I’m doing. So yeah, I think both of us two have pretty large, actually large language model in this case.

Rob Walling:
Cool. So let’s move on from there to the fifth point. And this is one I’m sure you’ve thought about. So I’m curious, I think we both agree on it, but the question here that I would ask myself as a founder, is my business model a ticking time bomb. Because some businesses are, if you’re a big team of analysts presenting data and you’re doing a bunch of manual work and you’re summarizing things like that becomes like you said, what was it the three-week report becomes a two-day thing? Well, once anyone can do that, that’s a problem. So how should founders, what should they be thinking about and how can you escape that? How can you not let your business basically be completely deprecated by AI in those cases? It’s not everybody, but there is a subset of SaaS out there.

Einar Vollset:
There are. And I think the use cases where you have… Again, maybe before it, a part of your mote was, “Oh, this is really dirty data. It’s hard to deal with. It’s imprecise. It requires verification.” So you end up outsourcing to a call center in the Philippines, or not call center, but outsourcing overseas, and you’re paying people to do textual analysis or updating reports manually because it’s super hard to do with the existing technology. I think that particular model, I think if that was my moat, if that’s what I did, I would be worried. I don’t necessarily think it’s like… So here’s the thing, it relates to what LOMs are bad at. You can’t have it. People say, “Oh, I can’t do math. I can’t do two times to the power, eight times, whatever, six, whatever.” That’s not like… Yeah, but that’s what it does.
That’s just a calculator. It doesn’t matter. That’s what it’s good at. What it’s also, another thing it’s not good at is queries like all the. Give me all the whatever it is in the world. It doesn’t do well at that just because that’s not what it’s good at. That’s not being able to iterate every role a database is not something that it’s good at. And so I think the moat that exists from building out and managing and quality assurance on a team of people in the Philippines say, I think that’s going to go away. But that doesn’t mean, I think that the entirety of those businesses are just three lines of code and ChatGPT anymore. It just becomes a different moat.
And if you lean then too heavily on that existing moat, then yeah, I think you’ll be out competed. But I think a lot, this is what I’m saying is for those kind of businesses, it makes sense to sit down and say, “Okay, let’s just admit that this particular thing is no longer a mode. What can we do given the capabilities that we already have and the customer relationships and the understanding of how we get that data, what can we build? It’s almost like, “Yeah, if you’ve been cruising along on that mode by yourself, you’re going to be in trouble.”

Rob Walling:
And I’m here less to be, what do you call it, doom and gloom? The world is not over. I mean, we are in as entrepreneurs, we are in the best spot possible. One of the best spots possible to take advantage of this. This is exactly, I keep likening it back to having a VA for $5 an hour because in 2007 and 2008, I was running these very small businesses. I didn’t have huge budgets, and I was either writing all the code myself and I was doing all the support myself because the only people I knew lived where I did in LA and I paid developers $75 an hour and I paid admins $30 an hour and I had no profit margin. And then when I read the four-hour work week, I was like, “What? I can do that?” And so I did, and it made my $3,000 a month net invoice suddenly be a 90% profit margin business. And that changed the game for me. And that’s when I realized, “Oh, I can do this.”
So I think AI is that plus, plus, I think it’s even more different. And I was in a position, and at the time people were like, “Oh no, offshoring, outsourcing, all Americans are going to have no jobs.” And I was like, “A, that’s not true. And B, a lot of us knowledge workers and developers and entrepreneurs are at a great place to take advantage of it.” So that’s my mental model of this.

Einar Vollset:
I agree. I mean, I think it’s an amazing time to be alive and it’s also an amazing time to be the bootstrap or nearly bootstrap a self entrepreneur because even just being aware that this exists and being able to write code, they can call an API, they can do some stuff that’s an amazing. You’re like heads and shoulders above your average, Joe and I think doom and glooming on it is completely pointless. It’s going to change. It’s change our lives, I think mostly for the better and having an optimistic view of where this is going to go and have it be like, “This is amazing. What cool things can I do now?” I think is the right attitude as opposed to the [inaudible 00:34:45] like, “Oh, all SaaS are going to go away. Everything’s going to hell. And then by the way, the AIs will take over and murder us all.

Rob Walling:
Right. And look, is there a spot to say, “Will people lose jobs and will it have social impact, especially on just the lower end of data processing, data entry, whatever?” Yeah. This is not the podcast where we talk about that kind of stuff, but it certainly could have an impact. So as a listener, hopefully that was helpful to hear Einar and my thoughts on this, obviously we have, I would say a relatively unique perspective being invested in a lot of companies and just being knee-deep in SaaS day in and day out. If folks want to keep up with you, you are Einar Vollset on Twitter, and we mentioned discretion capital at discretioncapital.com as well as TinySeed of course that’s what we work on together.

Einar Vollset:
Indeed.

Rob Walling:
Thanks again for joining me, man.

Einar Vollset:
Thank you.

Rob Walling:
Thanks again to Einar for joining me this week, and thanks to you for joining me every week. I’ll be back in your earbuds again next Tuesday. This is Rob Walling signing off from episode 663.