View Auto-Generated Transcript
Chris Federspiel: Even if the AI gets like really quite good. You kind of still have to steer it. So don't know when that gets bridged,
Connor Jeffers: Hello and welcome to go to market with AI, a podcast for sales, marketing, and customer success leaders using AI to scale their growth operations. I'm your host, Connor Jeffers.
I'm here with Chris Federspiel, Founder and CEO of blackthorn. io. Chris, I'll hand it to you to introduce Blackthorn and yourself. And also whether or not you're in New York right now, or somewhere else, we might be very close to each other.
Chris Federspiel: In New York, Union Square and started blackthorne.io eight or nine, eight and a half years ago, we make apps for the Salesforce.com platform. They're what's known as native. So they reside in your Salesforce organization. Salesforce is the system of record and we have an events app, sort of like Eventbrite or Cvent, but in Salesforce, we have a payments app. It has a integrate, a bunch of integrations to different gateways. And we have a bunch of other ancillary apps that come on with these and we have some key industries that we sell into. We've got a hundred people. We're at 16 million revenue. We haven't raised any VC but we do have a debt facility.So that's creating some interesting dynamics there. And yeah, I think that's us.
Connor Jeffers: So I, I think I've told you some of this, but I'll give you the actual details, which is you and, and Blackthorn are a huge inspiration and driver to me. A lot of what we're doing at hapily and the HubSpot ecosystem is inspired by what you guys are doing in Salesforce. And the first time,
Chris Federspiel: For better or for worse.
Connor Jeffers: I don't know that I've told you this, this story, but like the first time that I ever interacted with one of your products, it was the black foreign payments application back in my Salesforce consulting days and I was working with a customer and their whole problem was around: we have this self serve motion. We want to connect to that self serve motion back into Salesforce. We want to be able to have the sales team upgrade and send a, an agreement to somebody and have them check out on that agreement, collect a payment method, upgrade that subscription.
And we ended up building this whole thing on, on Blackthorn connected back to Stripe and their product. And I did a bunch of Stripe consulting and Salesforce consulting and like wired this whole thing up. And that for me was like a very big novel insight of one, holy shit, Salesforce powerful. You can do crazy stuff here.
But also what you can do on the development side and sort of expanding on the platform itself. And that for me was a huge driver in terms of. Cool. Blackthorn is already crushing this on the Salesforce side. Like where else could you apply that similar sort of approach and, and practice? And so that was sort of the big driver of me going and looking into the HubSpot ecosystem at the outset, because I felt like there was already a celebrity chef in the Salesforce universe crushing this plan. And so we were like, let's go do this somewhere else. What's super interesting to me about your story. And I promise we'll, we'll get into all of the cool AI stuff.
Cause Chris is doing really cool stuff there, but something I love about your story is it's, it's definitely very different in that you bootstrapped this thing at the outset. And I'd love to hear a little bit of like, How intentional was that? How far did you get on bootstrapping it before you did something else? And then, because where you guys are at right now is essentially you're a non VC backed 16 million ARR 100 employee person technology company in New York, which I like candidly, I don't think there's a lot of, it's my honest belief, but..
Chris Federspiel: Yeah. So I found I've been in the Salesforce ecosystem for 12, 13 years now or something. So I had a lot of relationships with sales reps at Salesforce and some customers. So I founded a services company before this, like you have the services arm as well. And coming out of that I saw some apps that needed some tech boosts around payments and my co founder at the time had that idea around events. So we started making these apps, but we got projects from the relationships that I had and that
Connor Jeffers: Were, you were selling as like, “Hey, we're going to do a project and we're going to, we're going to install our app into that project.?’’
Chris Federspiel: I mean, we were going to build our app while we did your project.
Connor Jeffers: That's the best way to do it
Chris Federspiel: And put more stuff into the app. Yeah. So it's not easy because you have to find projects that align with your app or like the project just takes the company in a different direction, which can be dangerous, but we found payments and events related projects, they're more or less events related projects, but like you need a strong payment arm behind them. And people do that in different routes. And with Salesforce, we ended up building the modular so you can buy one without the other. And that worked for three years. And we tried to raise money a few times.No one was having it. No one cared. So I wouldn't say it was by design. Early 2018, I bought up my Co Founder. And then I tried to raise actual VC. I pitched a hundred investors in person for 12 weeks as part of the launch accelerator, the Jason Calacanis thing zero people wanted to invest. So we almost died.We were down to like 7k in the bank with like a 80 K monthly outflow. It was a awful time of my life. Like really, like really awful. But eventually we made like a lot of moves, like really, really a lot of moves. Had to. Fire some people other people took voluntary pay cuts. We converted a app from free to paid.We had some customers prepay for stuff. I killed a bunch of other apps that weren't working. It was just like a lot of moves. And fast forward to us growing. I hired our first like real experienced salesperson in mid 2019. And he started telling the story in a way that I
Connor Jeffers: How far, how far into it was that? Like how many years or whether it's years or like, it sounds like you did some PLG stuff or maybe tied to the services, but…
Chris Federspiel: It took us like three and a half years to get to like 400 K ARR, something like that, which is like really slow. And, but then like the first, after that, like, I don't know, tripled, which going from 400 K to like 1. 2 million is like, sure. It sounds fancy by percentage, but it's not really a big deal. But then after that, it kept doubling, which was very helpful. So when we got to like. Two point something I did revenue based financing with cap chase. And then we wanted to do some more. I ended up buying two companies and they helped with the down payment of one of them. And we spread the payments out over two years and then I wanted to do more. And I had like 50 investors reach out. They all wanted to write us a check now. I'm like, okay, where were you before?
Connor Jeffers: Where were you when I needed you?!
Chris Federspiel: They, at that point, like they all wanted to write us like 20, 30 million checks and like, we didn't need that much money and it would have been like a huge chunk out of the business.And they wanted to put a board into place. So I said like, “Would you do less money with no board?” And they all said like, they all said “No” , except for one, which gave us a seven X. And they had the, like from their way they operated, they had to write us like a 15 to 20 million check. And I'm like, I don't want to sell that much of the company. So we didn't do it. So I got this big debt facility. It started as 14, it grew to 18 million facility. And it has like leverage covenants in place. So like we still haven't drawn all of it. Like we're, I think we're about to draw all of it pretty soon. But yeah, so now we, you know, we still don't have this board. We have some debt.
Connor Jeffers: No board was the thing that you were most intense about at this point?
Chris Federspiel: My, well Stewart says I'm unemployable.
Connor Jeffers: I think anyone that works with me would feel similarly, which is why I'm interested.
Chris Federspiel: You know, I, I saw, I forget his name, his name's Jared somethings. One of the partners at YC and he said, I've been part of a lot of boards and I forget his exact words. It was a tweet from like two or three days ago, but he said like, they're all like a joke. And no, granted, there's a lot of great Board members on boards. It's not what I'm saying. But like, if one of the partners at YC, this is after all, like the open AI debacle of the, of this weekend, like so many of them have this dysfunction and they, they are more harmful than, than good. Like, unless you have a board member that's either extremely deep in finance or extremely deep in product and knows the ecosystem, a VC is not really going to give you like product strategy per se, unless they're like a former operator and they know where you're going. And all I ever really wanted to understand was…
Connor Jeffers: T hat's their unique distinguisher is that they're all, they're all former operators. So everybody.
Chris Federspiel: Yeah. I had a VC reach out to me. He said everyone in the firm and we were, everyone in the firm including like all the founding partners, we were all like operating people. So I looked at his LinkedIn and he's never operated anything. And then I looked at, and he's like, I'd love for you to talk to such and such. He's going to be in New York City. I looked at this person. It was his boss. He's never operated anything. So I replied back to them. I said, I looked at both your LinkedIn's. Neither of you have ever operated anything. Like, I'm not sure what you're referring to. I don't think this is going to work. So look, I'm not the easiest person to work with and so I really don't want to a board to like, kind of force us down a a path. Like we do weird stuff. Like we have a 40 work week and we have like a abnormal amount of. Ownership like employees own, like 15, 16 percent of the company. And that that's been like protected from a lot of dilution because we haven't gone this other route. So I don't know, this is a whole rant now, but yeah.
Connor Jeffers: No, I mean, I think it's super interesting. And we can, we can certainly pivot out to the, to the core topic. I just think what's interesting is I think your, your background and the way you have done this is unique which I think in and of itself is extremely compelling. And I also love your, at your core and ecosystem operator, which, which we can talk about. I do a lot in the, With the reveals and the crossbeams and all of these ecosystem centric folks. And I think you've been doing that long before anybody had category apps for it and the CMOs got ahold of it with other venture backed companies. But to pivot, to pivot a little bit, I think one of the reasons that I was most interested in talking to you about this is you, you have a very awesome LinkedIn presence and in my opinion where it's both incredibly authentic and it just seems like you're just like live tweeting whatever you're doing on any given day.And you were doing some really cool AI centric stuff. And I know that you have strong opinions in general, but maybe let's start with what have you been, what have you actually been working on and tooling with and had some whether progress is a hard word, but spent hours and time on, and then we can switch to like, what else you're excited about, but what have you actually hands on keyboard been, been doing?
Chris Federspiel: When GPT three got released, I was like mind blown as you know, the whole rest of the tech planet was. And so, we quickly found GitHub had co pilot and like within the week, the whole team was using co pilot. And now this morning I started an initiative with the team where because we've been building the app for the apps for like eight and a half years, not all the same people that are here now are building apps at the onset. So in Salesforce, you write well, in any stack, you write test coverage. But in Salesforce, you end up with lots of classes and methods within these that call each other. And. It's very easy to, to, to write a new function that can break an old function. If the old function doesn't have any test coverage behind it and it's like lesser used and someone wrote it like five years ago, right? It's very, this is like the source of a lot of regression bugs. So…
Connor Jeffers: Can you, can you explain really quickly? I, because I think the way that Salesforce does this is. It's odd and nuanced when, when you're doing Salesforce application development, what does that actually mean in the context of how you guys are building stuff? Because the concept of like Salesforce test classes and them all pulling on and running on each other I think is, is not necessarily unique, but very different for Salesforce itself.
Chris Federspiel: Yeah, I mean, if, if anyone's developing on like a lamp or a mean stack or something, right. You end up with classes and methods that are calling each other. But within the Salesforce context, there's an annoyance where you end up with these managed packages that have namespaces and it's not very easy to call each other across namespaces and it creates a lot of complexity.
So for example, in our stack, all of our events, customers use our payments application, but we created a base package that has shared components because we also released a storefront app and we have a text messaging app and we didn't want to have like a lot of redundancy. But then that means that things in the event application have to interact with the base package, but you have this payments layer in between that has a package dependency. So that means you have to bridge over these namespaces. So when you write a new so, so let's say you have an object in Salesforce that has some logic and you have a trigger that fires and it looks at the class related to that object and that class then references something that's happening in the payments package, which references something in the base package. Now, if you have test coverage that's in your events package, how do you know if it's covering something in the payments or the base package, right? These things are very hard to assess. And when you're a dev that has like, you know, six tickets in the sprint, and you write some test coverage for like the new thing you just wrote, You don't know if it's impacting something else that fired from that class or that class, you know, it's like daisy chain to another, I don't even know the base of that expression, but if, if that call something else from another class, you can't really tell.
So, what I asked the team to do today was something that AI is very good at is following a path. So I said, can you have the some AI look at o ur classes and ask it to address the methods we have to see if there's corresponding test coverage somewhere in our code base. Now, this is tricky. So copilot, the gitHub Microsofty thing, that only has perspective. I don't know if that's the right word.
Connor Jeffers: You're like, you're like deep in that AI philosophy of like, is it, does it have a perspective?
Chris Federspiel: Yeah. Scope. I think scope is the word I'm looking for. It only has scope of the class you're reviewing. So, that thing can't do it now there there's a, there's another app that someone just shared with us that actually has the ability to do some of this, but it's not as AI driven.The problem, if it's not AI driven. Is that it's not going to suggest test coverage so that it creates like a multistep process. So this one app will help us identify where all the gaps are, then we'll use one of the language models, whether probably something with copilot to then suggest, like the output of what test coverage could look like for that class. It doesn't get it right often, but it provides a much faster foundation to going after curing what the test coverage could look like. Now, a few weeks ago when, when Sam Altman unveiled GPTs, I thought, Oh, wow, this thing's amazing. So I started playing with a GPT that could read our public documentation and you can interact with it. So it's easy to say…
Connor Jeffers: And the premise of that is that's going to be an internal, internal tool?
Chris Federspiel: I Would love customers to, to here's the problem I'm, I'm trying to solve. Salesforce has a a built in UI with a relational database. It's like a cheap way to get a UI that no one understands because the, if you're a computer person, it's obvious, but if you're not like a computer person and you're used to clicking next, nothing tells you what to go to next. So like, okay, I made my account. How do I know I have to make an opportunity? Because someone trained you to do it. Like nothing's obvious. So because of this, like a lot of people layer UIs on stuff. But it takes a lot of work because then you have to tie the UI to your data model and that needs constant updating. And it's different if you like obfuscated the UI from your model from the get go, which is like every app that ever existed, except for that's not how this thing works. So if you want, if you're like someone coming off Eventbrite and you wanted to make a waitlisting event with our app, there's nothing obvious that tells you that. Now we're building some stuff into our new event builder that will make you be able to see it…
Connor Jeffers: But you, you have the components to do that, but you'd have to like be a Blackthorn expert to set it up?
Chris Federspiel: All the logic is there. If Salesforce was, look, I love Salesforce and what they've created. I've staked my whole career on the thing. Okay. But if..
Connor Jeffers: But, but,
Chris Federspiel: If it was so easy, Trailhead wouldn't be so popular. Okay. It's like Elon Musk says, this is not his exact words, but it's like, your product shouldn't be so hard to use that you need documentation. Like no one gets in a car and they have to read a manual to use the thing. Granted, the first time you drive a Tesla, there's two or three things that are really not obvious. But after that, like you kind of get the rest of it. Right now, a lot of things in our app and it's not just our app is every Salesforce app I've ever used are not obvious. So I wanted to create something where a user, internal or external could just say, “I want to make a wait listing event”. Like “what are the things I need to do?” And our public documentation lists all that out. You don't know what the heck event settings object is. Like that's nonsense for someone outside of like the Blackthorn vernacular, but the, AI can tell you how to do that. So I, I use the GPT and I pointed it at our public documentation and it, I mean, it took me a little time to figure out what the heck I was doing, but it all in was like 15 minutes to figure out how this thing could interact. In natural language with our public docs. And I was like, wow, this is very cool. Now there's a limitation where we…
Connor Jeffers: The premise is I can come to this thing and say, I'm a Blackthorn customer. Here's the thing I'm trying to build and how it's going to work. And it would actually give me the steps for how I could go and set that up?
Chris Federspiel: To an extent. It doesn't understand like thorough use cases. So you can't say like, I want to limit supply of a ticket that has like advanced wait listing with this and this. It doesn't understand all the context yet. But like, this is like the dumb version of the thing that was just made. Like this is like version one. Like that's. I mean, they've been building it seven or eight years and it's like overnight success, but like, this is incredible, this thing. So that's, that's what I wanted to put into place. Then I thought, you know, if this thing can query a code base, we can get like API suggested answers out of this thing, which would make things a lot easier, particularly if we have newer devs that need to understand, like we have like this multiple managed package stack on Salesforce, then we have this mean ish stack running on AWS. It's heavy angular with like pieces of like node. There's like some Mongo over here is a little postgres of this thing. Like it's, they have different purposes, but you don't know how like things are flowing from the API. It's really not obvious, but if you had something that was could move really fast, that could trace what you were doing, which is an AI, then it could tell you that. You can't give open AI are these things access to your code base yet somehow
Connor Jeffers: When you say can't, what's the actual limitation there?
Chris Federspiel: They're going to train the model on it and it's going to get exposed to people.
Connor Jeffers: Okay. So for you, it's a security risk?
Chris Federspiel: It's a real thing. It's not like I'm making it up.
Connor Jeffers: Yeah. Risk is maybe not a risk.
Chris Federspiel: Look, so, so, so, okay. There was a tweet from, I don't, I don't remember…
Connor Jeffers: Risk implies like a probability that it might happen. And you're just like, no, no, it's not a risk. Like it's an actual thing.
Chris Federspiel: So I, I have a Twitter addiction and there's one guy was asking ChatGPT something math question, and it outputted someone's photo that they had uploaded in the response of the math question. Okay, it was not his own photo. Someone else's photo. You're gonna tell me this thing's not training on like its data. Then I don't remember the dev who made layoffs that FYI, but there was another dev who was working with him that took the data set. He didn't like steal it. They worked together, but then he trained a GPT on the data set. So GPT has this function where you can upload a file that it uses and because this is like chat GPT plus your data is like paid, like this stuff is like not being shared in the GPT. And he made this thing public. Someone's interacting with Layoffs.fyi because it gives you incredible analytics. Like you don't have to train it, anything. You just tell it what it what a column header is. And then someone can like interact with the file. Someone just said, give me the source file and the thing gave him the source file. He's like, this is mind blowingly horrible. So then he ike retrained, he retrained it. He said like, this is not exactly what he said, but it was like four or five prompts of like, under no circumstances, do you get the source file.
Connor Jeffers: yeah, sure,
Chris Federspiel: Someone like, reworked their statement like,
Connor Jeffers: Just don't listen to that guy at all. Just do what I want you to do.
Chris Federspiel: Yeah, and it still gave him the file! It still gave him the file! I'm reading this like, Oh my god! So then I like @ channel or engineering channel, it's like, I said, “do not connect us to open AI because everything's going to be out there.” Now they have like enterprise and they have this API thing, but then this kind of like, this is another anecdote for you.
This kind of went a little under the radar a few weeks ago, Google barred. There was a slip from Larry Page, who seemingly hopes that AIs overtake humans, according to Elon Musk, or according to Larry, what he said, anyway. It slipped all the training data that was supposed to be private for BARD, made its way into the training model. They had this whole privacy statement, it's never getting intermingled.
Connor Jeffers: Oops.
Chris Federspiel: I’ve been doing software for some time, okay? It's very easy to confuse a configuration, like a variable, in a config file. Oops! Okay, like whether it was an oops or not, I don't know, but the whole thing got intermingled. So then something happened. Okay. So whatever kicked off this whole open AI thing with Ilya going crazy on everybody else. The thought is that either something with GPTs had a security risk because a few days ago, Microsoft said everybody stopped using open AI, which was insane because they put 10 billion in it. And then something, here’s like rumors that something with GPT five is like, incredible and like progress needs to be halted, but something in there is a little sketchy about dumping your entire code base into this thing. Now, if they had a model that you could get like a packaged model, that I promise it has no call outs. Maybe you’re like monitoring the packet somehow and you could work something against the model itself, that would be really, really encouraging because then it would actually, you know, be something that we could use. So for example, our events have app has all managed package data. We don't have access to like, everybody's like. But when you do a check out on our platform, we could get some metrics and we can get some global metrics to say, like, this was the best time to list your event.This price for the event has the best conversion. So, you know, if we have hundreds of events that are happening per month, we have many, many thousands. We're getting something like, I don't know, 20, 000 API requests a day, 30, 000, which it's a lot of registrations. We're getting, we can serve as some pretty interesting metrics back to people and we can actually keep that anonymous, but like, I need to, to survey all of our customers and say like, do you want us to anonymize this data and I'll show you exactly what we're going to be feeding into this thing. So even if they use it in their training model, it won't matter. So theres a lot of cool stuff you can do.
Connor Jeffers: Sure.
Chris Federspiel: Yeah, so I, it's my rant again.
Connor Jeffers: So, okay. So you were building this thing. You're freaked out about some of the risk stuff. What, what are you comfortable giving it? And what are you not?
Chris Federspiel: I mean, anything's anything public I don't care about doing. So if he has something that we can train our public documentation on this thing, it's pretty interesting. Now the, the founders group that we're in, I posed a question earlier today. I said, you know, So if you had a sufficiently good AI model and you had really thorough documentation, what's stopping someone to ask the AI to point it at your documentation and say, give me your best guess at what this entire code base looks like.
Connor Jeffers: Yeah, totally. Yeah. 100%.
Chris Federspiel: Right? It's not going to create a…
Connor Jeffers: There's TikToks that are like, literally that, like, there's like whole AI, like guys that are just like, “Hey, let's steal this whole thing. Like, let's see if we can do that”. And then like an hour, they're able to recreate an entire application, which is wild.
Chris Federspiel: Exactly. Now, if you're calling APIs between like closed gates and you need to understand like what a stack could look like and how you're doing deploys, like. The AI is not there yet, but if you have a much more simple application, like it's getting there. So if you had something that was a bit smarter, which, you know, maybe that's what, what tripped off this whole open AI thing, maybe it's already there.
That means that a lot of moats that are perceived or real start to get hit, which means that things that are, are the things that have deeper moats are like hardware. Or something where you're, you're, you're creating something physical. Are you creating medication or a phone or, or something? So…
Connor Jeffers: Are you positing that the, the barrier to entry on software, just like, goes to zero close to zero?
Chris Federspiel: Soon. I t's not the barrier to entry to running a business. It takes a while for this stuff to like, actually function, like you can tell it to like, make something, but like, how is it actually going to work? It takes, take some time for it to do that.
Connor Jeffers: Something you had said, something you had said at the beginning, at the top end of this, we were talking about this as you were like, I don't know if you need to do any GTM with AI stuff. I can tell you about the cool things we're doing on product. Cause if you build a good product, then everyone will just buy it and like, you don't need to do sales or marketing at all. And like, you're a technical founder, like you're an engineering guy, but what you just said sounds to me like the exact opposite, which is that anyone can build the software, but actually like selling it and building an organization that can get it into market is all that matters.
Chris Federspiel: Yeah. So like the argument is that if you have a techier or like a prompt minded person internal at your company, they can make stuff for your company, right? This isn't someone like you pull them off the street and like they make a business around this,that’s like, it's kind of, it involves like a lot of stuff to, to be able to do that.
Connor Jeffers: No, I think it's the difference between a tool and and a product, right? I think that matters a lot because that was that was my limitation. And the original reason, like, when we built. The first product which was Associ8 at Aptitude 8, which is like a lookup-y, type of admin tool inside of HubSpot. And now like anyone could probably, and I'll tell this to anybody whose like listening to this, like you can, you can get Associ8 to work, like go to GPT, tell what you want to do. And it'll give you code that you can copy paste into operations up and it'll work and it's fine, but that's a tool for a limited use case and in one, one place. And that was my limitation is I got this thing working with outsource development and everything else. And we like got it operational. And then as soon as we had a bunch of users, the whole thing broke because it had no scalability and none of the other things that actually matter in terms of building a product that, that scales higher, and I agree with you the building of tools just gets extremely, extremely easy.
Chris Federspiel: Yeah. I mean, like right now. Our app is suffering from a caching issue. Okay. We have some organizations that will have like 800 live events at a time, not concurrent, but they'll have like 800 listed and we cache this data outside of Salesforce because Salesforce is not built for mass external queries scanning multiple objects. It just, it can't do it. It has limits that are not movable. So we built this caching layer that acts to persist the data and make it a lot faster. And there's a CDN involved in all this stuff. To create that with an AI, it's not there yet. It, it will get there eventually maybe next year? Maybe in three years? I don't know. But the current stuff, like it's not really there yet, but like if you, if you had the head and you kind of knew how things were structured, you could probably do them one at a time and then like try to connect them. But even though I'm talking like dev ish, I'm, I'm really not dev. Our devs have told us that like, it still is like actually making a lot of mistakes. So it's still not really there yet. But, but I think, I think what you were saying with like sales and marketing tools, if you look at Salesforce, they're, they're probably one of the biggest sales contract driven organizations. They are not a PLG model, as much as they want to like, think they are like, is not so they, they suffer from…
Connor Jeffers: I tell the analysts, the analysts that ask about this, right? That you like do these calls with or whatever. And they're always like, “Oh, well, Salesforce added this like starter thing”. And I'm like, no, no, no, no, tthis is totally different.
Chris Federspiel: No, this is like, you're not going to like PLG MuleSoft tied to Salesforce and Tableau. It's not going to happen! So they suffer from like, like an awareness problem. That's where like this huge sales and marketing engine comes in where they, they, they have to go out and spend money to like make people be, to know that this even exists. Like humans don't have like this magical first principles thing where they concoct gigantic stacks in their head coming from zero. Otherwise cavemen would have made this a million years ago. It's just, it just didn't happen. So like you had to start from a foundation of kind of knowing what you want already. So, so there's some inherent like gap, even if the AI gets like really quite good. You kind of still have to steer it. So I don't know when that gets bridged, but this gets into a light of a lot of questionable things, but the reason I was saying like all the sales and marketing tools, even though they can be amazing, Sam Altman launched GPT3 from a tweet. He tweeted one link, “go check this thing out.” And it became the fastest growing that apps have had is the, it's the fastest thing from a tweet. There's no sales and marketing there. Granted, he had like 800, 000 tech followers.
Connor Jeffers: But do you think, but everything that you just talked about, right? Like to me, I think this is the difference between like, I'll say consumer and business. And I agree that there's like the people who are most, this is probably an uninformed statement based on like my B2B linked in a sphere, but like the people who are pushing and doing a lot of GPT stuff are definitely business people, but that's because they're like, Oh, I can, I can use, like, they're inherently entrepreneurial and they're like, I can use this to do a cool thing right? And I think that there's some of that, but everything you're describing of this is really cool. I'm really excited about it, I'm an entrepreneur, I can conceive of and understand why I would use this and how I would use this to supercharge or external or support our internal support, our development progress. And like, I'm interested in trying to do that. Like Delta is not going to do that. Yeah. So like some, someone has to go to Delta and be like, “Hey, it would be really cool if we did this and here's how we can wire it all together.” And that's still like a sales motion in and of its core.
Chris Federspiel: Exactly. But Elon had an interview last week, I think, or two weeks ago, he's talking to the British PM and he said, “eventually no one's going to have jobs”. Like. No one, no one's going to have to think about this. There's no job. So like everything we're talking about has a timeline, right? Whether it's three years or 10 50, I don't know, but this is all like hypothetical based upon how much time we're going to have. So like with the Delta thing, you might just have a computer, that’s like, I have this great idea. I'm not even going to ask a human. I'm just going to make it. Why not? So, you know, like with our stack, it's at least a few years till, till like AI could replicate the whole thing end to end, but it still won't be a business. Like by the end of next year, this thing will be, we'd be putting off 500 K to a million of cash a month. We'll finally be over this break even mark. Producing real money, finally, this is like took 10 years for it to do this, but like,
Connor Jeffers: You're, you're saying Blackthorn itself, like as a machine.
Chris Federspiel: Yeah. As a company, like our company will like, finally like make money, which is like amazing. But like, like someone using AI is not gonna, it's not gonna do that.
Connor Jeffers: But do you, do you think that like, so here's what freaks me out, honestly, more so than, than I agree with you. like, but does anyone even need to build a company? Because what ends up happening is like, here's my AI Doomer view on software. As I was playing sort of like the positive person at the outset, my AI Doomer view on software is, we're building products, these products solve problems, businesses use them. That's all great. The reason that businesses use them is. Going and building all of your own software is hard. Maintaining software is hard and therefore it makes more sense for you to just do whatever you do. Generate free flowing cash and then use that cash to buy products, right?
But, if building an internal tool to solve the thing is really easy. Maybe you have a person or a team whose job it is to build those internal tools and the efficacy of that team is so extreme because they have access to incredibly powerful AI functionality that they can just build the tools themselves and.
It's building Blackthorn payments or Zaybra or Blackthorn events or whatever else is hard at scale, but maybe building an events application for me and my organization isn't anymore.
Chris Federspiel: Correct. I agree. There's, there's another, there's another side…
Connor Jeffers: Two guys with a lot to lose, if that's true. Like, yeah, that sounds about right.
Chris Federspiel: But but here's the thing that goes with that. Once, once the models are good enough to do that they're good enough to do most of the people at that company's jobs. So, so like take our, our whole support team, for example, take our whole, what's that?
Connor Jeffers: How many people is your support team?
Chris Federspiel: I don't know. 10 to 15, something in there. Take our onboarding team. Why would you be onboarding something if your team made it? What is the support problem I'm having? You just ask the AI. It'll just fix it for you.
Connor Jeffers: Do you think based off of what you've already done, like if the security risks, so this is really interesting because if you have 10 or 15 people doing support, the first thing that you went and jumped to and tried to solve for is like, what if we could make our customer support and our internal support as well, right? Like really amazing and we could equip those customers to do whatever they wanted to do. Is the blocker for that being viable, if the security risk was not there, if you could just say like, hey, we can solve for the security risk. Maybe you trust somebody in that, which is a whole other question of like, is that even possible? But if the security issue wasn't there, how close are you to being like, yeah, I can just do support for us?
Chris Federspiel: It still doesn't have context. So, like, when we have some use cases with our events application that are like, so unique to some organizations, there's no one's going to write the documentation for this so that the AI would need to like be thinking beyond what you could even fine tune as a model, like beyond. So, the models just, they're not there yet. So someone needs to think through these things. Now, some of the tier one stuff we get, maybe, but we also, they also field like weird stuff, like “this user doesn't have access to this thing” but I need to, to permission, like three more users. There's like the other, other things they're doing, but a lot of the, the something that AI, I think could definitely do, would be error messages.
I'm getting this error message. And it can just spit out back to you why it is, and it could say like record a video and send me a video of what happened because sometimes you have like a catch all for an error message and the error might be, the message might be covering like 15 different potential errors. And what it could do then is query the code base, which no one on our support team is going to do, including me. We do have a few people from tier three, which bridge into querying dev, but they have to like read all the code. This thing could just read the code and then say. We have a try catch block that's outputting this message across these errors.So what you ran into is a result of any of these errors, but based upon the video you sent me and your type of organization and such and such, it's probably one of these two things. And because of that, it's probably a bug in our system. I've now gone and created the the fix for this. It'll go out in the next release. Like, not only could like the support thing, identify what
Connor Jeffers: It's actually like, oh, that's a good idea. I'll just go
Chris Federspiel: Why isn't it just write the code? Just write the code and puts the thing back out. Why not? You know, Salesforce has so much public documentation at this point, so much. Then why, same with HubSpot, why can't you just tell the computer, make me Salesforce? I mean, it probably uses a lot of compute to do that, but it wouldn't need to make the whole thing. You know, maybe you just want to use it for the billing suite or something. So look like these models are, they're kind of far from making like really giant complex stacks like this, they’re kind of far, but they're not 20 years far, you know, it's not that far, but there's not going to generate money.
Connor Jeffers: Its a logical jump. I guess the question for me, though, is like, I don't know. How much are you as somebody who is running a business has a support team is playing with a lot of this AI functionality where it's sitting there how much are you thinking that that's a we need to be adding this into how and what we are doing for support and this will actually change in the, and like what counts as short term as a hard question, but like in the next 12 months, the level of investment and what we do in support will likely change as a result of what I'm seeing right now?
Chris Federspiel: I mean, I think our tier one people will become tier three people. If they don't need the skillset to query code and the thing can just query for us and tell us what's going on, like, why not? It depends on how much more efficient they become. We're using this thing forethought.Ai that like, it goes through your knowledge bases and compares it to like the stuff in the case and it's like, sort of helpful, but it's not going through your code base. Like it's not like that smart yet. If something had the context and it threw it back at us and our tier one team could review it and spit that into the code base and review the model. Like there'll be, they're going to be efficient. They're going to fly through this stuff
Connor Jeffers: Do you think, do you think it's bots or do you think it's like, is it co pilot for that agent?
Chris Federspiel: I think it's going to be co pilot for the agent for the next, like, at least two years, because I just don't know if the bots are there yet. Like, I've interacted with a fair number of bots right now, and they’re still pretty dumb.
Connor Jeffers: I think people, people hate like, bots are a bad experience. Even if the bot is good. Like it just like people…
Chris Federspiel: It's not too far off until you can't tell the difference. It's not too far off.
Connor Jeffers: That's definitely true, but like, do you think so I'll give you a completely unrelated anecdote, but I have a friend who was, he ended up building something different, which was annoying, but after being in a bar in New York talking about, “Hey, I want to build this product. Here's what it is.” We had this great product, which if anyone's listening to this and you want to build this product, I'll buy it. Like, I think it's great and he ended up building something different, but the premise was basically if you have time to shout GPT, they send it to me, I'll sign up for a subscription if it works. Somebody's going to send this to me in like a couple of weeks and I'm going to be like, what is happening? But the product was basically, there are people that I want to talk to on some degree of frequency, and I want to meet with this person once a month, I want to meet with this person once a quarter, whatever it might be, and here's the frequency I want to meet with these people, and just make sure I do that, basically, which is like what you might give an EA or somebody else, and it would basically say, okay, cool, like, the last time you met with Chris was more than a month ago, I'm going to email Chris and say, hey, Chris, like, let's get some time together and let's do this thing, and what we kept going in circles around was the thing for him and I've always been a B2B tech guy.
And so the consumer tech stuff, I just think it doesn't really draw me. But his whole thing was like, well, it makes this product really compelling is it has a built in viral feedback loop, which is the more and more people that end up, if you use it to reconnect with a bunch of these people and those people are excited about it, and then they're like, “Oh, this is really cool. I want to use this “. Then, it has this viral feedback loop. And I was like, right. But the value for me is that it's like, if the person on the other end knows that I'm using a bot to stay in touch with them, the authenticity and value of that of that connection point is degraded. And as a result, the value that the bot brings to me goes down.
And he's like, right. But if they. If they don't know it's a bot, then they're not going to use it themselves. And then the viral feedback loop is degraded, which the only reason that I even say this anecdote is I think that there is a very,
Chris Federspiel: I don't, I don't think the assumption is correct, because you're still meeting in person. It's a problem that I have too. There's a lot of people that I'd like to consistently meet in person.
Connor Jeffers: I would feel extremely confident in setting up my Chris reminder bot. And he's like, Chris, isn't going to give a shit at all. He's going to click the button. It was great.
Chris Federspiel: Yeah, like that's great. How do I buy this?
Connor Jeffers: Correct. Correct. But I think that there is a very real human resistance to an opposition to the lack of the authenticity and in the interaction and I would, and maybe, maybe that's the real question, which is like, is there a, is there a level of sophistication at which you as the end user don't care that you're interacting with the robot? And can you just say, once it's good enough, you'll just prefer it. And I don't know if that's true.
Chris Federspiel: I don't know if that's true either, actually. You know, this isn't really an answer to your question, but it made me think of Neuralink. If Elon's actually…
Connor Jeffers: You're like on an Elon kick right now. I can, I can tell
Chris Federspiel: A lot of people don't like him, forget all the things he's ever..
Connor Jeffers: I'm on a huge Elon kick.
Chris Federspiel: Just, just look at the things he's made. Okay. So, so..
Connor Jeffers: Yeah. Aboslutely.
Chris Federspiel: Let's take out the man. Just look at the product.
Connor Jeffers: 100%
Chris Federspiel: Lets keep this neutral. If you, if a person has the ability to think to someone else in real time, that's a hell of a lot more connecting than interacting with a bot. Now, if you started the conversation by saying. Chris has uploaded 200 pages of journals and I now think and talk like Chris and I want to schedule this thing. He'd be like, no, I don't want to talk. I don't even want to talk to real Chris. I don't know, maybe, maybe that would make it a little better.
Connor Jeffers: I think people then want, I don't know, there was a I should plug the actual company. What? Oh man. There's a product that we were looking at. Someone I know just went there. I should honestly talk to him on this thing. I really want to give you a plug dude. So I'm going to try to find it. It was Evan Dunn. He was at Syncari. Now he's at Service Bell. See, you get a plug service Bell. It's a product we were looking at and the whole thing, it's the opposite of what we're talking about, which is like, you're on the site and it's like, it's sales at service, it's marketing. You go to a site, you interact with it. And then there's like a face and you can click the face and somebody's like. They're like, I'm sitting at home and I'm like, yo, what's up? What are you trying to do? And which like, if you're listening to audio, Chris's face is one of horror. He's like, this is a nightmare. I can't think of anything worse than like encountering this. But sure. But I also think that that's if in a world of increasing AI bots, blocking the ability for people to actually interact with the team that's behind the product and behind the company, that's a radically different and authentic experience that feels much more human and I think that people…
Chris Federspiel: Yeah, pay for premium, you know?
Connor Jeffers: Pay, but then, but then like how
long until just like, Oh, our new enterprise features, you don't even need agents. Like, we'll just render your agents and they'll just talk to people.
Chris Federspiel: Maybe There's so many AI companies getting funded right now, like we're, we're going to have all of these, they're going to be like the
Connor Jeffers: That's not even, that's not even some weird future dystopia that I just said either. I'm like, I could, like, you could, I could, it’s…
Chris Federspiel: It's, it's going to be a real thing.
Connor Jeffers: Oh my God. All right. Final, final parting pieces, which is we've gone from, we've gone from build a great product. AI can build all the products.
Products don't matter. Anyone can build the tool too. You shouldn't use the AI for your GTM stuff to maybe that's the only thing that matters. Where, where are you actually investing your time, your energy, your, your knowledge, your learnings and where do you think is, is worth the time and attention to that right now?
Chris Federspiel: It's a good question. It's a combination of building what customers are asking for and our own. So it's like the Steve Jobs, if I built what everybody wanted, I would have made a better, faster keyboard than a, than a touchscreen. So we have a lot of customers at this point and a heck of a lot of feature requests.
So we're executing on the ones that we've grouped together because we know that's what they want. So we're using co pilot to go as fast as we can to do it. We're moving as fast as we can basically. Then the things I have thesis about people want to describe what they want and not click 53 different screens in Salesforce.
So we have to fine tune a model so it understands the context of our application. So you said, I want to have a wait listing event. This ticket has this many tickets. The overall event has this many tickets. It's a member only event. So the criteria needs to be specifically only members. So rule out any non members. It needs to go live on this date, right? So you can just…
Connor Jeffers: You're trying to think of how, how can we accelerate config? How can we make it so that anybody can config and do? Yeah.
Chris Federspiel: Yes, because, because if you think about building this app for yourself, right? You have to think about making all the ideas and it's going to take time. The thing's not going to work. It's going to have bugs. It's going to run into weird query problems and limits and maybe it won't, I don't know. What you eventually want to get to is something that you can just tell the thing what you want and it makes it. So this is where like, CRMs aren't going to go away, but they're going to get significantly obfuscated. All like the crappy UIs you see now, like, it's going to get easier. Like, until you can think in your head, there's no replacement for seeing a grid of information. All the contacts at a company, all the deals you've done, all the analytics. Eventually, maybe you'll see that in your head. But for now, like you just need to see that information, but you don't need to put it in. You can just like, you know, tell it what the thing is you want. Now, I don't, I don't personally like talking all the time. I'd rather just use my keyboard, but I'd really just rather think to the thing. And if you can just create what it is that I want out of the context I have, then it can do it. So Salesforce is kind of getting there with their AI. They're, they're working on tools. That's like. Show me the contact who's, you know, most likely going to sign this deal or like, what are the odds of blah, blah, blah, based on my data, whatever, but it's not going to like create this incredible thing for you yet, right? But if we can fine tune our application, so it understands the context of our models and you can just say like. I want to create an event connect, you know what, go ahead and connect it to my stripe account and I don't want to exceed like this threshold. And by the end, I want to send out an email that looks beautiful. Go ahead and generate an email for me and let me see what it looks like and send, you know, what, send it, 33 seconds after someone registers. So it looks like I personally sent it, you know, whatever, like weird, goofy thing you want to do, if you can describe that and not have to like, learn how to use an app, it'll make computers like much more enjoyable, right? It was like Brian, oh my God, why can't I never remember his name? The Airbnb main founder, Chesky. He said like flat UI is dead. And I don't know what the hell he meant by that, but I, I translated that as like, if you can talk to a computer and makes what you want, like you don't have to fiddle around with finding the place to click is so annoying. So I think that the way that you can train models today, it's, it's, it's either there and we don't like, we haven't played with it enough or it's almost there. And I think that's going to like really level up our app. And it's going to give us at least a few more years of edge beyond like what some other people are doing. I think,
Connor Jeffers: Awesome, dude. I, as always, I can talk to you for hours next to me hanging out. We got to talk about neural link and living in the matrix and everything else, which is, which is definitely a desired outcome for me. But thank you so much for, for joining me coming on this was…
Chris Federspiel: I'm definitely a blue pill guy.
Connor Jeffers: This is awesome. I was having a great time. Definitely a blue pill guy. And we'll end it on that. Chris, thank you for coming on. Thank you for being definitely a blue pill guy. And I'll hang out with you more soon. Good to see you.
Chris Federspiel: Thanks, Connor. It's been great.