The 5 Levels of AI Coding & the Collapse of Human-Written Code with Trevor Hinesley, Co-Founder & CTO at Soundstripe

(Intro Narrator at 00:00:00) Today, we're welcoming back Trevor Hensley, co-founder and CTO at Soundstripe, to talk about the five levels of AI-written code. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:16) I have learned in the preparation for this conversation, I have learned something new about these five levels of AI coding. Can you help me wrap my mind around that? Let's talk about these five levels. How did you find this?

(Trevor Hensley at 00:00:32) Okay, great question. I found it through a YouTube video. The guy's name is Nate something. I can't remember. I just like his stuff. It's kind of easy to listen to. It's basically, like, level zero is fancy autocomplete, right, of using AI. So, you know, for anyone that's listening that uses stuff like Supermaven or when Copilot was like autocomplete, it writes that. Level one is when you kind of graduate from autocomplete to you are using AI to assist with like boilerplate code. And that could even be like you're asking it for something and you copy-paste it out back into a file or whatever, right? So that's like, regardless, you're relying on the AI to write more of the code. Level two is where most teams are at. You know, you can take like the classic Quad Code experience where like you're sitting there, you're running it, and you're going back and forth with it. You're reviewing all the code that it writes, et cetera, but the model is writing majority of the code.

(Trevor Hensley at 00:01:34) Then level three is where the model is writing all of the code, and I think this is where you start to basically, you're just reviewing the output. Like, you're less concerned about like micromanaging it as it's doing a task, right? Level four is where you go so far as to it is just feature spec in, so not just like a task or something, but like a full feature in, and then you don't look at anything until it's output, like, ready for PR, right? And then level five is you turn the lights off in the factory and you go home and it just runs without you. So that's the dark factory. And it's a very interesting way to look at it. And, I mean, frankly, I think everything is headed to some form of level five. We've been having lots of conversations about this in my company.

(Joel Beasley at 00:02:26) Yeah. Well, look, I'd agree with you. I think that I think there's going to be an expansion between four and five. Like, I think there's going to be more levels added because—

(Trevor Hensley at 00:02:38) I could see that.

(Joel Beasley at 00:02:39) Like, I would say I'm pretty confidently on, I'm like two-three right now on my personal projects.

(Trevor Hensley at 00:02:46) I'm just reviewing, making sure it's doing the right stuff, and—

(Joel Beasley at 00:02:50) Long term, it won't be causing relational database issues, things like that. And it's so good. It is so—and it gets so much better. It gets—I know. Trevor, we are here. We are definitely here.

(Trevor Hensley at 00:03:07) Could not agree more for numerous reasons. That to me has been the biggest unlock, I would say. And plenty of people talked about this, I guess, like, in the broader zeitgeist, but just the fact of like all the projects that my ADHD brain would like come up with that I was like, "Ah, it's going to take six weeks. Like, it's a cool idea, but I'm not going to do it." Like, the other day, have you ever used those like online diff checkers where you just like drop paste in—you paste in text on both sides of a thing, and it'll just show you where the diff is, right? Okay. You can—there's like, you know, there's like Linux or any kind of Unix system has like a diff tool. You can do it in terminal. I just never liked it, so I'd always just drop it into diffchecker.com or whatever it was. But sometimes it's like—like I was dealing with something that was like a secrets file, but it's a big one. And so I didn't want to eyeball which of these random string of characters was different, right?

(Joel Beasley at 00:04:03) You just wanted to paste all your secrets—

(Trevor Hensley at 00:04:05) Exactly. But I didn't want to paste it—yeah, I didn't want to paste it into diffchecker.com either. So all I did was threw up a—I used Superset. If you're familiar with that, it's kind of like Conductor, which is another one.

(Joel Beasley at 00:04:17) Okay. Okay.

(Trevor Hensley at 00:04:17) It's—if you haven't checked it out, you should. It's better, in my opinion, than Conductor. It's basically like Conductor, but it's terminal native. So everything is just terminal windows. It's not like a fancy UI, which can honestly—Conductor may be in some trouble because Claude Code is now starting to detect whether it's in a UI, and they're going to prevent subscriptions from being used in those tools. So anyway, I was using Superset. I just threw up a new project in there and was like, "Just make me a Mac app that's diffchecker.com, but on my computer." And in like two minutes, it spat it out. And I use it. I use it all the time. Like, I never would have spent the time to do that before. But like these little things like this, I am having a blast with, and it's a prime example of like, I didn't need to review a lick of that. Like, I just needed to be able to see if there's a diff between two pieces of text. And I'm never going to touch it again. It just works, you know? That to me is a very tiny example of, you know, if you extrapolate out the concept of like, today is the worst the models are ever going to be. I can do things like that on very tiny things that don't really matter much. But like, if you do extrapolate that out, at some point, I should be able to do things like that on things that really do matter, right? As the model gets better, as we get better at sort of restricting them, like, putting guardrails around them, that kind of thing. So that to me has been the most exciting piece of—I can also like, when I'm doing something like this, Superset can be running like five different projects at a time, and then I can just go review them when I'm done, answer a couple things that it's waiting on me for, right? Or like, give it some corrections, go into the next set of meetings, and then it just keeps cranking away.

(Joel Beasley at 00:06:11) Okay. You've got to help me understand. I don't know Conductor. I don't know Superset. What is this?

(Trevor Hensley at 00:06:16) Okay. It is—have you used Git worktrees at all?

(Joel Beasley at 00:06:20) No.

(Trevor Hensley at 00:06:20) Okay. So Git worktrees are basically a way where you can have, you know, when you clone a repo from GitHub to your machine, right, it's in that folder. You can be on one branch at one time, right?

(Joel Beasley at 00:06:31) Correct. Yeah.

(Trevor Hensley at 00:06:32) Git worktrees allows you to have multiple instances of the repo on your computer all in different branches, so you can work on multiple things at the same time.

(Joel Beasley at 00:06:39) Got it. Okay.

(Trevor Hensley at 00:06:41) Superset and Conductor are basically tools that automate setting up Git worktrees. So like, I can hit the plus sign in a project, which is a repo that I've set up, and it will create a new worktree, which means it loads it onto a separate folder on my computer, sets up a separate branch for it. And then I just put in a description of whatever it is. I can even put in a prompt, and what it will do is it will take the prompt I put in. It will translate that into some meaningful branch name. It will spin up a branch in a worktree and all that. It will throw up whatever agent I prefer, which is Claude for me, and it will drop the prompt I initially put in there into Claude and put it on auto mode and just let it go. So I can be working on five different features on one app at the same time, and what's cool about Superset is it has like a little ding that happens whenever the model stops and needs your input on something. So I can just see when these things are at different points where I need to go give input or whatever. And it also has like a little summarize thing to where I know what it was just doing. It's amazing.

(Joel Beasley at 00:07:47) So this is like—so I'll tell you my stack. My stack right now is I use Cursor.

(Trevor Hensley at 00:07:51) Yep.

(Joel Beasley at 00:07:52) And then I have the paid Cursor. And I'm building an application, and it's been great. We wanted to save—like, we had this application, like $500 a month, and we were using 1% of what it could do. It was basically like cap table distribution stuff.

(Trevor Hensley at 00:08:07) Got it.

(Joel Beasley at 00:08:07) And so we were like, yeah, it's just like, I just did it as an experiment. If you're like, if this were—I spent a weekend. I was done with it in a weekend. And now we're just like making it cooler and better. But that's what my workflow would be, is I would describe it and I would instruct it to like continue to solve problems and stuff until it achieves the outcome because that was one of the cool things that I first experienced with Cursor that I could do. And then I would go downstairs, go for a walk or something, come back five, ten minutes later. It's got like a couple questions for me. I answer the questions, and then it goes works for five, ten minutes or whatever. And so the idea of just doing that but parallel makes complete sense because you would essentially create all these different feature branches of non-conflicting feature sets, right? Like, I need the authenticate—you know, whatever. That was probably the worst example. I chose the first example—I chose the one system that touches so much.

(Trevor Hensley at 00:08:59) Right.

(Joel Beasley at 00:09:00) But like just like some new report, some new model that needs to happen, something, and then you instruct and it creates it in different feature branches essentially and then you can like merge them back into each other. Is that what's going on?

(Trevor Hensley at 00:09:13) Yes. Because when you get to the end of any particular—like, let's say you've got five of these things and they're all working on different worktrees, blah blah blah. It can just put a PR up with that branch, and then you can merge it back into main or whatever. So what's interesting about it is it's not different from if you have five different developers working on—

(Joel Beasley at 00:09:34) You're right.

(Trevor Hensley at 00:09:35) —branches, right? It's exactly the same thing.

(Joel Beasley at 00:09:38) Local different copies. Yeah.

(Trevor Hensley at 00:09:38) Yes. You can just do it yourself. And, again, the level of autonomy you give it or whatever, like, that's totally up to you. You can be looking at all of them if you want. But it's been really effective for me because I can still get, you know, a lot of the meetings and stuff I need to do done while also working on work that I never would have had time simultaneously to do.

(Joel Beasley at 00:09:58) Here's a question for you. People who have brought up Claude Code a lot.

(Trevor Hensley at 00:10:02) Sure.

(Joel Beasley at 00:10:03) Are they—is there some separate application? Are they just choosing Claude Code as like the model in Cursor? Like, how are they—when they're saying "I use Claude Code," what are they saying for real?

(Trevor Hensley at 00:10:12) Okay. It depends. Like, I use Claude Code in a terminal. So like, I don't—I don't use Cursor.

(Joel Beasley at 00:10:19) Okay.

(Trevor Hensley at 00:10:20) Yeah. A lot of people are doing that, but there are a lot of people that—

(Joel Beasley at 00:10:24) A lot of people are doing that. So you don't use an IDE? What do you use in your terminal? Like, what is—like, I understand, obviously, the terminal.

(Trevor Hensley at 00:10:31) I do not look at files anymore.

(Joel Beasley at 00:10:34) You just don't look at files anymore.

(Trevor Hensley at 00:10:35) I don't. We have a policy within our company that there is no human-written code ever.

(Joel Beasley at 00:10:43) Oh, snap. Yeah. Oh.

(Trevor Hensley at 00:10:44) Yeah. It's—oh, okay. And we—so we, like, I look at code when it's done with something, but I just do it through the lens of like, I can—Superset, I can like click a file. It'll, you know, it'll show like the—you can read it—terminal output and stuff and whatever. I can like click a file name that it's working on and it'll show it in a viewer or something, but it's not an IDE. Like, I don't—I literally do not use an IDE and haven't—

(Joel Beasley at 00:11:09) So it's like a chat interface and terminal? Is it your actual Mac terminal app and you've installed something with Brew or, like, what is it?

(Trevor Hensley at 00:11:17) So it is a term—Superset is a Mac app, and it has like, when you're looking at like a window—

(Joel Beasley at 00:11:24) So that's the application you open. You open a hundred app—okay. So you're not in Terminal, like, the very specific application on Mac.

(Trevor Hensley at 00:11:31) Yeah.

(Joel Beasley at 00:11:31) Oh, okay.

(Trevor Hensley at 00:11:32) Yeah.

(Joel Beasley at 00:11:32) Okay.

(Trevor Hensley at 00:11:33) Yeah.

(Joel Beasley at 00:11:33) Okay.

(Joel Beasley at 00:11:36) Okay.

(Trevor Hensley at 00:11:36) It has—I think it's actually using like xterm.js where like—okay—it will load a terminal for you.

(Joel Beasley at 00:11:42) Yeah. Claude has, or I'm sorry, Cursor has a terminal inside of it.

(Trevor Hensley at 00:11:45) Same same same same same same.

(Joel Beasley at 00:11:46) Oh, okay.

(Trevor Hensley at 00:11:47) Got it. Yeah. But it is not an IDE.

(Joel Beasley at 00:11:49) But it's not an IDE.

(Trevor Hensley at 00:11:50) It is—I guess it has some like, it has like a little changelog sidebar that's kind of like a Git commit thing. You can see what's staged and stuff, but it is just not an IDE. Yeah.

(Joel Beasley at 00:12:02) I'm looking at the screenshots now in GitHub. That's pretty cool.

(Trevor Hensley at 00:12:05) It is cool, and most of my team is using it at this point. Some people are using Conductor, which is another one. Most of us are on Superset though. Like, I really prefer Superset. I think it's kind of—

(Joel Beasley at 00:12:15) So you would say like IDEs are going to be like if you're code-writing first or you're just a legacy caveman and like, and you just have to have that feeling. But looking at the interface of this, this is like the interface that would be designed if the way you are core running your development is these multiple agents, multiple things, non-human writing code?

(Trevor Hensley at 00:12:35) That is correct.

(Joel Beasley at 00:12:35) That looks like what the interface is designed for.

(Trevor Hensley at 00:12:38) They actually have said they're trying to get Superset to where like, if you wanted to run a hundred agents at once, the interface should be intuitive enough to not get lost doing that.

(Joel Beasley at 00:12:48) Correct.

(Trevor Hensley at 00:12:49) Yeah.

(Joel Beasley at 00:12:51) Well, I don't know. I feel like an old person now because I'm like, Cursor's fast enough. It's fast enough for me. You know, I can—yeah. But, you know, what's cool is the Cursor—I don't know if you know this, but like, you can't—you can't—I don't—I'm speaking way out of turn here. One of the things I have found cool is you can't have two conversations like running in parallel in Cursor.

(Trevor Hensley at 00:13:12) Really?

(Joel Beasley at 00:13:12) Yeah.

(Trevor Hensley at 00:13:13) So how does it function different than what I just described?

(Joel Beasley at 00:13:17) I'm not smart enough to answer that question. It's not meant for—I know that I've talked to it like in two different threads like going at the same time.

(Trevor Hensley at 00:13:27) Okay.

(Joel Beasley at 00:13:27) But I have never—my brain is not—I'm learning all this new information today. I didn't even know this was a thing until just this conversation. This like people developing with multiple agents at the same time type deal.

(Trevor Hensley at 00:13:41) Okay. Yeah.

(Joel Beasley at 00:13:41) This like—on the same—like, my brain understood it like if you have multiple projects. Sure. On the concept of like a single Git repo situation, that's new. This is all new for me right now.

(Trevor Hensley at 00:13:53) It is antithetical to how the human brain works. Like, we are not really wired to be able to context switch between, like, for instance, five different features at once and remember, like, okay, where was—what was this thing working on? It needs my input now. Like, that is frankly, it's exhausting. I'm not going to lie. Like, that is actually mentally taxing, but the productivity boon is undeniable. So like, I've just had to accept that like, we've got to figure out a way to do that, you know?

(Joel Beasley at 00:14:28) Yeah.

(Trevor Hensley at 00:14:28) And—

(Joel Beasley at 00:14:29) Should hire more women.

(Trevor Hensley at 00:14:31) Seriously, dude. Seriously.

(Joel Beasley at 00:14:33) Sounds like having a conversation with my wife.

(Trevor Hensley at 00:14:35) That's what I'm saying. My wife is a thousand times better at me than that. But it is, that part is taxing. But like I said, we're so, like, I'll give you another example. We wanted to do a tech stack migration recently because, frankly, this is my hot take.

(Trevor Hensley at 00:14:50) I hate JavaScript. And I am trying to do as little JavaScript as possible. We are trying to do as little JavaScript as possible as an organization. So we're trying to do a migration out of React. So we built a bunch of automation that allowed us to, we ran Codex on a 24/7 project for over two weeks autonomously.

(Trevor Hensley at 00:15:18) So, like, we would just check in every day or two, and it was very expensive token-wise, but it was an autonomous execution of a complete tech stack migration. And it wasn't just we threw a prompt into Codex. It was like, we want to move to this tech stack, good luck. It was we spent some time for a couple weeks building both deterministic, like bash scripts and things like that, and probabilistic. Right? So stuff like skills, things like that. To build an autonomous, I would say, like, add-on capabilities to the harness that is Claude Code, in this instance, Codex, to where we could run this autonomous thing, and we'd be confident that it was going to be checking its work. It was going to be verifying screenshots to ensure the CSS it's doing matches what we expect, that it actually is looking at acceptance criteria every time it finishes something, that it's doing its own refactoring passes, it's doing its own reviews. We ran it for about two and a half weeks. This was a migration project that we had specced out pre-AI to take about a year and a half of three to four people's time.

(Trevor Hensley at 00:16:26) Big migration. Right? It did 50 to 60% of it in two weeks. And so now we're polishing and finishing it now.

(Joel Beasley at 00:16:35) Josh. Josh. Can you jump in? Can you remind me who's the guy that's, like, doing this? The, I love them. We had on their CEO. Blitzy. Oh.

(Trevor Hensley at 00:16:46) Oh, I haven't heard of them.

(Joel Beasley at 00:16:47) Oh, Trevor. I, they, I got reached out to by them, and they're like, yeah, we want to, they did some sponsorship stuff with us and everything. Like, we want to, we're making this claim, and they, I go,

(Trevor Hensley at 00:17:01) no.

(Joel Beasley at 00:17:03) I'm not putting that out there. Like, that's not a, so I said, I said, here, I need to see a demo. I need to talk to the founder. Like, this is all presales before we took him on as a sponsor. And it, it's, it's what you described, but on steroids.

(Joel Beasley at 00:17:15) And it was like a 99% rate on the first draw. Like, you run it and then, like, it, it almost is fully perfect. It, it, unbelievable. Blitzy. It's called Blitzy.

(Joel Beasley at 00:17:28) Blitzy. Yeah. Yep.

(Trevor Hensley at 00:17:30) Okay. I am, I'm looking at that as soon as I get off this.

(Joel Beasley at 00:17:34) Yeah. Yeah. They and they charge by, like, you know, like project or something. I don't remember. But I do know that if you tell them that I sent you that they give me some money or something, which is great.

(Joel Beasley at 00:17:44) Okay. But that doesn't change the fact that the interviews, you could you can watch the two interviews, one with the founder, one with his customer. I interviewed his customer, and I think those are the two interviews we did with them. But basically, it's not a waste of your time. It's what, it's what you described is probably, like, their first prototype ever of what they've done, but now they've gone just, like, insane with it.

(Joel Beasley at 00:18:11) And and they're just they're growing so fast, and they're, it's crazy. Yeah.

(Trevor Hensley at 00:18:15) I'm so interested. Yes. I'm 100% going to go check this out after.

(Joel Beasley at 00:18:19) Okay. Now, I want to talk to you more about level four and level five. Because we have level four, feature spec in. You know, you don't look at anything till you're ready for PR, like, type deal. And level five is, like, the dark factory. Okay. Well, like, where's the part where, like, we enter, like, the customer changes the app. Right? That, that, like, so where's that part? Because, like, you know, you, you can interact with the customer and build the feature spec. But, like, if you really are going to turn the lights off and go home, something has to be engaging with the customers, making decisions of what to build or direct, or, or was that turn the lights out?

(Joel Beasley at 00:19:00) Was that just a little bit, like, tongue in cheek? Like, we, we still going to be directing it to some degree?

(Trevor Hensley at 00:19:05) It's, it's a good question. So when I think of, I, well, frankly, I think that could look many different ways depending on what the company's intent is behind the dark factory. Right? But the idea is you would not need oversight to whatever degree makes sense, I guess, for your use case. Because, I mean, one example is, okay, just let the whole company run itself. That's a, that's a full dark factory. Right? But when I think of it like that, I'll give you a, this is what we're working towards internally. We have this concept for a tool called Backstage, because everything we're doing is music themed, so we're calling it Backstage.

(Trevor Hensley at 00:19:38) And it's basically a very simple, you can think of it like a chat interface, like an LLM, any typical, whatever. But the idea is someone nontechnical, like a sales rep in our organization, could be on a call with a customer. Customer can have a pain point that we don't currently address, but we think is something we can handle. They could, I mean, you could even automate this part. Right?

(Trevor Hensley at 00:20:01) It could take the meeting transcript and start working on something that could do that, or the sales rep themselves could prompt the system. And this system would be the dark factory where we have set it up to do everything to the coding standards we want, to do the proper review, to do a deploy preview for whatever this is, et cetera. They're on a call with a customer. They basically prompt a system with, they're nontechnical, like I said, they prompt a system.

(Trevor Hensley at 00:20:28) It creates a prototype of whatever they have prompted it with, with a spec that is to our standards, all this stuff, spits out a prototype before they ever get off a call, and they can show a preview of it to the customer and say, is this kind of what you're thinking? Whatever. When they're off the call, they could bring that over to engineering because it's a deploy preview, it's a PR, whatever. And then engineering can effectively take that through to deployment.

(Trevor Hensley at 00:20:53) Right? Or we can automate that piece. But the idea is we can have as many humans in the loop in this if we want. We don't have to. So we have control over what our quality levers are, what our touch points are, and where we want to have checkpoints in the system, if any.

(Trevor Hensley at 00:21:09) What's really cool about that to me too, beyond just the fact that that's crazy that we can even talk about this and that be a possibility today, is what this enables is engineers effectively, their job goes from being the executor for product development to being the executor for building systems that enable anyone else to do product development within the org. Because right now, like, you know, engineers and these, the worlds of product, engineering, and design has, like, long since gotten muddier and squishier and more sort of collapsed. Right? But I think what's interesting about this is that engineers get to focus on pure systems building in this paradigm that we're talking about. Right?

(Trevor Hensley at 00:21:58) You are building systems that then enable other people to do the things that prior only engineers could do. That to me is super empowering and interesting. And it, it goes beyond just engineering. It would include codifying things like, you know, instead of just dropping the brand guide into a Claude skill and hoping it works well, it would go into, like, building a system that actually codifies and follows to a T the way we use the brand guide or the way a designer uses our design system, which is more than just here's the design system. Right?

(Trevor Hensley at 00:22:36) These kind of things, they're a little squishier. Right? But this is what, to me, what that, what systems thinking for an engineer looks like in that world. It's not that engineering is gone. It's that it is focused on building systems that enable more than just them to do that work.

(Trevor Hensley at 00:22:54) Kind of, it's just a different, it's a different spin on, you know, prior, they were building systems for customers. Now they can build systems that allow others to build systems for customers.

(Joel Beasley at 00:23:06) That's the future.

(Trevor Hensley at 00:23:08) I think so.

(Joel Beasley at 00:23:09) That, that's, that's, that's the in between the dark factory. It's going to be, like, a concierge service. Like, I call up, I talk to the, the salesperson, they prototype through the conversation or whatever it may be, then an instance of that prototype, you know, gets kicked back to the, the client, and they're like, yeah, that would solve our problem. And it's like, great. We have that. And then from there, it's a business decision of do you integrate that? Do you let your product go in that direction?

(Trevor Hensley at 00:23:34) Exactly.

(Joel Beasley at 00:23:34) And, and if you're doing this at scale, if you're a large company like an IBM or Microsoft, and you're doing this at scale with hundreds of products and thousands of features, you can then start to look at, okay, in the past week, a variation of this feature has been requested 75 times.

(Trevor Hensley at 00:23:51) Yep.

(Joel Beasley at 00:23:52) By customers that make up 30% of our revenue. Implement, deploy. Yep. You know, because, because it's so low cost to have that salesperson in that sales call, you know, sales engine, we'll call sales engineer. Sure.

(Joel Beasley at 00:24:06) Like, in that, in that call to kind of, like, figure it out, that if you just have enough data and you have enough of these calls, it'll tell you where the product needs to go without one single person, like, having to meet with all the customers.

(Trevor Hensley at 00:24:22) Exactly. I also think you can, if you take that even further, right, there's the, there's the whole, you may have seen some of these, like, concept videos that were floating around of, like, generative UI where, like, this one, I say it's interesting, it's kind of a dumb concept, but it was a, it was an incarnation of what this could look like was it was someone using, like, a Finder app. Right? So, like, a file viewer type thing.

(Trevor Hensley at 00:24:50) But the app had no UI. It's just whenever they would click a folder to go into the folder, it would build a UI specific to that folder's contents every time they did that. So the idea was, like, instead of having static UI that is prebuilt, you could have systems that, like, build a user interface around the context of whatever is currently being done, around the user's preferences, around the data that is in context, whatever it is. Again, the example was, like, kind of arbitrary. It's like, why would you need that to be generative?

(Trevor Hensley at 00:25:23) But the point is there is a world where you can take that same system and make it available to customers to basically build their own, they can do it themselves. Right? I don't know if I love that because people can ruin their own experience that way. Right?

(Trevor Hensley at 00:25:40) Like, it's the same reason you don't implement all customer feedback. Right?

(Joel Beasley at 00:25:44) Well, that's a good point. And when I interviewed the creator of the Internet guide, the, the World Wide Web years ago, like, probably four or five years ago, he said this is kind of where we're headed, and I'll be more descriptive here. He said everything's going to turn into, like, a, like, like, your bank account is going to turn into more like a database.

(Trevor Hensley at 00:26:03) Yep.

(Joel Beasley at 00:26:03) And the, it's going to be kind of irrelevant what interface is being used, this customer service's interface being, and everything's going, like, to start looking like a database. Like, right now, we go to Facebook, we go to Instagram, we go to things where they also, it's, they, they control the interface and the database. Right?

(Joel Beasley at 00:26:22) Because, and if that's going to decouple in the sense that your interfaces are going to exist and then make changes with the databases based off of permissions. I was like, that makes so much sense. He said that's where the future was headed. And he said this five, before AI came out, before all the AI as we know it today.

(Joel Beasley at 00:26:40) He said this five plus years ago.

(Trevor Hensley at 00:26:42) Didn't you say that was, was that Tim Berners-Lee? Who was that?

(Joel Beasley at 00:26:46) Uh-huh. Yeah.

(Trevor Hensley at 00:26:46) Okay. I was about to say, well, he's obviously got a pretty good mind for predicting where things go.

(Joel Beasley at 00:26:51) But at the time when he was describing it, I was like, yeah, okay. Like, like, I didn't, I didn't disagree that it would happen, but I just didn't see, I just didn't see it the same way I can see it today because today, like, I, I even wonder how many of these businesses are even going to exist because, yeah, is it a tough ask for me to single shot a HubSpot clone today? Yeah. Is it in three years?

(Joel Beasley at 00:27:15) Probably not. Right? Probably not. It just might be the shared knowledge. Eventually, someone's going to create models that have shared knowledge of how all these systems work and clones of them, and they're not going to have to start from scratch to clone Salesforce or HubSpot or whatever it is.

(Joel Beasley at 00:27:31) They're going to, it's going to know the, the framework for that. And then we might end up with all of our own personal instances of everything that network in the background somehow. You know? Right?

(Trevor Hensley at 00:27:43) I, I, dude, I think that is certainly a potential future. I would ask, what do you think happens to software products at that point?

(Joel Beasley at 00:27:55) Honestly, I think the technology is like in Dune where it all kind of fades into the background. And we start, be focused, and it's going to create a huge identity crisis and economic instability on the in between. But I'm not talking about the morality of those things. I'm just thinking about where it might go.

(Joel Beasley at 00:28:13) Sure. And I think that's where it's going to go. I think there's going to be an increase in in-person events. I think we're going to have this magical little rectangle that'll pop up anything we want to see anytime of day, which it already kind of does. Sure.

(Joel Beasley at 00:28:25) But I think it's just going to do it in a way where we're, we might not have so many applications as what, like, because eventually, like, how many applications do you have on your phone versus ones that you go into? Very true.

(Trevor Hensley at 00:28:38) Yeah. It's a...

(Joel Beasley at 00:28:39) Small percentage of them. Right? I'm always like, how do I have 700 applications? I use, like, three things. Three.

(Joel Beasley at 00:28:45) Evernote, mail, and text, and phone.

(Trevor Hensley at 00:28:48) Like, I...

(Joel Beasley at 00:28:48) Don't, I don't know how I have all that. But you do, but eventually, you're not going to need that. You're just going to say, hey, Gemini, I want to, you know, I want the thing. I need, I've got this need to do the thing. Here we go. It's just going to be this, like, conversational situation, and it'll just show you what you want to see. Like, I want to be, Trevor, I want to say, did you see the Instagram video going around where they took, they made an episode of The Office where Michael, like, battles AI, where, where he battles Claude? Dude, perfect.

(Trevor Hensley at 00:29:25) Really?

(Joel Beasley at 00:29:26) Perfect. I don't know how much time they spent on it. It was like a three minute, they just did like the intro and like the first hook of whatever the episode. It wasn't a 30-minute episode. It was like a three to four minute thing, but shot for shot. I, I, I thought when I was watching it, not lying, I thought that they got together and did this as a skit just because AI had become such a big deal.

(Trevor Hensley at 00:29:48) Wow.

(Joel Beasley at 00:29:49) I thought, I thought, I thought they all got back to, you want to watch it?

(Trevor Hensley at 00:29:52) Yes. I want to watch this.

(Joel Beasley at 00:29:53) Alright. Is this it? Yeah.

(Joel Beasley at 00:29:54) Yeah. That's it. Oh, yeah. I did it with audio off, so I don't know. Everyone meet Claude.

(Joel Beasley at 00:30:01) He's an AI, which stands for absolutely incredible. Corporate sent us a subscription. Michael bought it a chair. Claude is here to make everyone's jobs easier. Watch this.

(Joel Beasley at 00:30:15) Claude, first task, research this. That's a stapler. Can Claude track, kill, and dress a mule deer? No? Then I'm not worried about it.

(Joel Beasley at 00:30:31) Alright. That's pretty much it. Come on.

(Trevor Hensley at 00:30:34) That, I'm not going to lie, that's gold.

(Joel Beasley at 00:30:36) Yeah, that's the voices, that's the almost magical fan-favorite parts of how the transitions happen with the classical scenes. They did it all, and it was like perfect.

(Trevor Hensley at 00:30:50) That is honestly one of the best. And maybe it was too, like I wonder how much of that was AI written, right? Like the dialogue and stuff like that.

(Joel Beasley at 00:31:03) Because that's—

(Trevor Hensley at 00:31:03) There was a—

(Joel Beasley at 00:31:03) Whole post about it. Yeah.

(Trevor Hensley at 00:31:05) Oh really? I'm gonna go look that up after. Yeah, that was my—

(Joel Beasley at 00:31:07) But yeah, for me—and Josh has heard me say this too many times because he's the editor here—but I always tell this story. I say, look, I've been saying it for like three years, right when AI first came out. I had this vision. I was driving home from pizza. I picked up the pizza, had the kids in the car, and I was driving home. And I was in the car with them, and I said, "One day, guys, we're gonna pick this pizza up, and I'm gonna be able—" It was like Christmas time, like a Tim Allen movie. Like I wanted to see another Santa Claus, like a new one. I'll be able to type that in, hit send, and by the time we get home, twenty-minute drive, by the time we get home, that movie will be ready to play and it'll be perfect. And it'll be exactly what we hoped it would be. Because that's going to happen. You're going to be able to say, "I want to see a comedy with Adam Sandler and Tim Allen, and I want them to be buying and selling and flipping car dealerships." Like, I don't know what we would call it. And then just boom, you get home and it's got the cover art on it, it's got—you know—and you just hit play and you just watch it. And when we get to that point, that's going to be a really weird point, because we will literally be able to do anything our minds can think of as far as creative. And at that, oddly enough, I think we'll put the phone down more. I think we'll put it away more, because it's like, why? Like, I know it's gonna be perfectly timed. I know it's gonna be great. I know it's gonna be whatever. And then sometimes you'll be like, "I just—I'd rather not look at it." Right? I'd rather just go.

(Trevor Hensley at 00:32:40) What do you think about, on the flip side, is that dystopian?

(Joel Beasley at 00:32:48) I don't know.

(Trevor Hensley at 00:32:49) I mean, is it like—do we—you know, I think I think about this a lot, especially given, you know, my—we have a belief within our company that, because like I said, we're super AI native, like very forward there. But we also believe in human-made music. That's what we—that's basically what we're selling. Right? So we think about this a lot. We're licensing it out. We're not selling it. But we believe that even if you're doing generative work, right, like if you were to throw in a human-made song, we believe it elevates that work because there is something intangible about what we can put. And again, I mean, there's—you know, this has been argued ad nauseam as far as like, are the things we deem intangible just a big data problem? Right?

(Joel Beasley at 00:33:41) Yeah. Yeah.

(Trevor Hensley at 00:33:42) And that's a totally valid conversation. But there is something interesting about, you know, even the nostalgia we have for our favorite bands or songs that remind us of certain moments, and then we connect with whoever wrote that song's story. And so it just has a different meaning to us because it's human to human. Right? So take the Santa Claus thing you just gave. Maybe that would be true, but would you have the same connection to that new Santa Claus movie as you did to the first time you saw it? And was it because you saw it in a theater with your friends or family and that's why you have a connection? Or was there something about the fact that you'd been socialized into liking—was it Tim Allen? I think it was. Yeah.

(Joel Beasley at 00:34:23) Yeah. Yeah.

(Trevor Hensley at 00:34:24) I don't know. This is—it's an interesting thought experiment. I don't have an answer for it.

(Joel Beasley at 00:34:28) Yeah, no, I know. And you're my guest here, so I don't want to blow up your whole thing, but—

(Trevor Hensley at 00:34:34) No, dude. I want this.

(Joel Beasley at 00:34:35) Yeah. Yeah.

(Trevor Hensley at 00:34:35) Yeah. Yeah.

(Joel Beasley at 00:34:35) Yeah. So look, I think it's hilarious how you are one of the most advanced technology leaders that I've had on the show as far as pushing the boundaries of what's possible with your company and improving with the levels that we've discussed. Sure. But then on your home page, it's like "human-made," like 100% human-made music. Now look, will it exist, and will there be a premium for it? Yes. There are still people today who have horse farms that make millions of dollars.

(Trevor Hensley at 00:35:05) Of course. Yeah. Yeah.

(Joel Beasley at 00:35:06) They have carriage services that make millions of dollars and are international businesses. They still exist today. Sure. You can still go ride one through New York City. But in relation to how many cars there are—

(Trevor Hensley at 00:35:17) Yeah, 100%.

(Joel Beasley at 00:35:18) It's like—so I think the hope side of me says, if you do it very well and you're a master of the craft, there will always be work for you. Yeah. So that kind of lets you sleep at night. That being said, 95% of the industry is going the other way. Dude, it's—

(Trevor Hensley at 00:35:36) Which is also true.

(Joel Beasley at 00:35:37) It is true, because when I think of—I mean, you can look at even some of the top songs on radio right now are AI created. Right?

(Trevor Hensley at 00:35:45) Are they really?

(Joel Beasley at 00:35:46) Yeah. I don't listen to—

(Trevor Hensley at 00:35:46) Two in the—if I can't remember specifically—there were two of the top songs, whether it was streaming or some obscure Billboard chart or whether it was the Hot 100, I don't know. But there was like two Gen AI songs that I think were in some top 10. And so it's happening. But it's also like—okay, to take another example of—like, do you remember when Girl Talk came out? This was a DJ.

(Joel Beasley at 00:36:17) No. You ever heard of Girl Talk?

(Trevor Hensley at 00:36:17) Okay. So Girl Talk is like a mash-up artist. So Girl Talk would take like—he'd take like a Jay-Z song and then mash it up with some like eighties dance tune beat, right?

(Joel Beasley at 00:36:30) Okay. I know what you're talking about. Okay. Yeah. Yeah. I just Googled it. Yeah.

(Trevor Hensley at 00:36:33) He was like what I remember of as—he didn't—I don't think he necessarily started this, but he was probably the biggest, like, mainstream hit at the time. And this was when I was like in college, like '08, '09, something like that. And like, there is a world for that, but it also didn't—and I mean, you could take this any route, right? Sampling, EDM music, like stuff made from a laptop, right? Like, it didn't get rid of everything else. It actually expanded musical possibility and stuff, right? And similarly, I 100% think there is a home for Gen AI music. Like, I actually—I mean, especially as I think about stuff like—like there's just—I can think of real use cases where it makes a ton of sense.

(Joel Beasley at 00:37:18) Elevator music. Exactly. Right. Done.

(Trevor Hensley at 00:37:21) 100%.

(Joel Beasley at 00:37:22) Never pay another artist. Just whatever the compute processing is, it streams it on a loop infinitely. Elevator music.

(Trevor Hensley at 00:37:28) Right. But will it—you know, to take like maybe a hyperbolic example, will putting an AI-generated version of a Beatles song that's not actually a Beatles song, if that ever became legally viable, if that was put into, you know, some epic movie scene as opposed to putting "Eleanor Rigby" or some other actual Beatles song in there, are they gonna have the same connection with people?

(Joel Beasley at 00:37:56) I think you're bringing up a great point because we don't talk about this enough. We all talk about technology as if we are all in the same generation and that exact same generation will persist indefinitely.

(Trevor Hensley at 00:38:13) 100%.

(Joel Beasley at 00:38:14) And that's not how humanity works.

(Trevor Hensley at 00:38:15) Works. Agreed.

(Joel Beasley at 00:38:16) We're multiple—crossing multiple generations every couple years, and then they all die off, right? Well, for now. Yep. And so I think a lot of the magic is—like for example, you know, you've got young kids. My kids are growing up with the fact that they can just ask Gemini to generate a song and it generates a song. Like, yeah, there it is. There's the song. Yeah. They're gonna be emotionally attached to it the same way we were emotionally attached—I was—to Blink-182, the first album I ever purchased, right?

(Trevor Hensley at 00:38:46) Sure.

(Joel Beasley at 00:38:46) And so like, that to me, and when I hear it, I've got the memories, I've got my childhood, I've got the bedroom I share with my brother. Like, it's all there, right? You know?

(Trevor Hensley at 00:38:53) Right.

(Joel Beasley at 00:38:53) Right.

(Trevor Hensley at 00:38:53) Right.

(Joel Beasley at 00:38:54) And so when this next gen—when these generations come up that are happening right now—they're just gonna enter a world where Gen AI is a native thing, right? Just like we were almost native on the internet. You're around 40, right?

(Trevor Hensley at 00:39:07) Yeah. Yeah. Yeah. Yeah. Well, I'm 36, but I'm close enough.

(Joel Beasley at 00:39:10) Yeah. Okay. Yeah. And I'm 38, so we're not far off at all. So like, we just take those things as just part of our reality.

(Trevor Hensley at 00:39:21) Totally.

(Joel Beasley at 00:39:21) We're not debating that realm. We're just like, this is the infrastructure for the internet's here, and we're gonna use it. Absolutely. But if you look up, when we were in our twenties and stuff, the people in their fifties were questioning the morality of it, and like, "Oh," they were all arguing about if it will exist. But we didn't have the argument at all. We just were like, "This thing is cool, and it's here. Let's build the next generation."

(Trevor Hensley at 00:39:45) Sure.

(Joel Beasley at 00:39:46) And so I think that's happening right now. We're all talking about like, "What's AI gonna do in the next generation?" Like, who cares? Like, this is awesome, and we're gonna do it.

(Trevor Hensley at 00:39:52) Right. Yeah. Yeah. I completely agree. And I think that that is—that's also, like, in our case—on that same line of thought, I guess. Ours is not because—like I said, I think that there is a world—I think Gen AI has a place. I use it all the time. We use it in the business for operational stuff. And my take on it is that it's kind of like what you were talking about, right? I think there is a premium on human-made music for what we don't know how to quantify yet. Like, I believe that in my soul that there is something different about a song that someone blood, sweat, and tears crafted over, that they put some sort of output that is intangible into it. Right? Now can I pinpoint that? Can I show on an audio waveform how that is distinctively different necessarily from AI? Not today. But I also think that regardless of—I don't think it's actually a sonic quality, for example. Right? Like, the sonic quality of a Suno track is there. Right? Like, it sounds amazing if you've ever messed with Suno or whatever. And even Gemini, like you said, like Lyria and some of that stuff, those sound amazing. But it doesn't have the—even the outside-of-the-box thinking that a human would have due to their own experiences, due to whatever. And now, again, can we get to a point where you can try to codify some of that into a model or an agent or whatever it may be? Certainly. But it still doesn't have the things we don't understand about human consciousness. And so like, even if we get to a point where we do understand that, just like we buy vinyl records when we don't have to, right? I think that there is a value on the things we do that connect us to one another, which is expression. Right?

(Joel Beasley at 00:41:55) Yeah. Where my brain's going right now is we have physical bodies. We're operating in real time.

(Trevor Hensley at 00:42:02) Right.

(Joel Beasley at 00:42:02) And we have experiences, and we're processing that in a way that the LLM can only process our scraps. It can only process if I go tell you about the story and then it tries to infer. It can't yet go out and have these experiences in the same way that we can. I think it would need a physical body to do that or something.

(Trevor Hensley at 00:42:23) Maybe.

(Joel Beasley at 00:42:23) I don't know. Maybe I'm outdated. But, you know.

(Trevor Hensley at 00:42:28) Well, even if it had that, right, it's like, does it—it's not even a question, I guess. For me, it doesn't mean—like, for instance, I can't go to a show, for example, and see if it was like an AI that was doing a show, right? Like I said, it's like a DJ set or something. Yeah. I can have a particular kind of experience there. It can be incredible. I cannot have the same type of experience of going to a show and finding my new favorite band because there is a culmination of the people on stage, the heart behind that particular performance, what they—you know, if it's their own songs, like, what was behind them writing that, like, maybe even their image. Right? Like, we've all talked about forever how that affects the brand of a band and stuff like that. There is value in sort of the holistic side of that, just like there is value in a Coca-Cola can beyond just what their formula is. Right? Like, their brand means something. And what is that? Can we quantify that? Tough, you know. It's super interesting to me, though, because to your point, I am on both sides of this conversation.

(Joel Beasley at 00:43:35) I know. I love the human stuff. And, you know, in our generation, I am the type of person that I tip the musicians. When I'm at the—when I give my kids the money, they put the money in. Most people don't. I do. When I see someone that's trying to live out their dream and trying to make it, I'll throw $10, $20, whatever. I just try to signal that like, "Hey. I see you out there making it happen. It's hard." Sure. And I think that for me as a consumer from—and there's a certain amount of me's out there. Until I die or whatever, till we go into the computers, I will continue to seek out and—if we're in a world where 80%, 90% is Gen AI, if I see a human thing come up, I'm still gonna throw a couple bucks at it. I'm still gonna support the human thing for the pure fact that it's human-made. Yeah, I will. What will it be as a percentage of my spend? Probably very low. Sure. But I will still do it. Now, will the generations two or three deep below us do it? Probably. It's probably gonna come full circle. It'll be really cool to have Gen AI, and then it becomes lame, and then it's cool again. Just like—yeah.

(Trevor Hensley at 00:44:49) Yeah. Just like vinyl records. Again, that to me is like—I think I can't remember the stats, but I think vinyl record sales are like—they just kept going up for so long. Like, it was—

(Joel Beasley at 00:45:00) Well, when they hit zero, it's easy to go up from zero.

(Trevor Hensley at 00:45:02) 100%. No doubt about that. No doubt. I mean, it was gone for a while there. But it's a substantial market at this point, whereas CDs are gone. You know? Like, that's gone. Very interesting. So there you go. Look at that.

(Joel Beasley at 00:45:18) That's exactly how I felt it was.

(Trevor Hensley at 00:45:21) Totally. Now, this probably doesn't take into account inflation, but what is crazy to me about this is that that is basically half of the peak. That's no joke that it got back to like half of the peak.

(Joel Beasley at 00:45:31) But if you had population adjusted, it's probably really little.

(Trevor Hensley at 00:45:34) True. Good point. Good point. And again—

(Joel Beasley at 00:45:36) Inflation, I'm sure. Because the seventies had people going like bunny rabbits.

(Trevor Hensley at 00:45:42) That is hysterical.

(Joel Beasley at 00:45:43) That's how they had it going. All right, dude, I want to talk to you forever, but I do have another thing I got to go to.

(Trevor Hensley at 00:45:50) Oh, it's all good.

(Joel Beasley at 00:45:50) I know. This has been fantastic. I am—so first of all, I want to apologize. I rarely talk this much in an interview. You've just—you were so exciting. It was so much fun to talk about this stuff with you.

(Trevor Hensley at 00:46:00) Dude, I appreciate that. I love, well, one, I just love coming out here and chatting with you. But two, this topic is my freaking favorite thing right now, especially all the dark factory stuff. It's just so interesting to me.

(Joel Beasley at 00:46:11) Trevor, we made a podcast. How do you feel? I feel great. Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it.

(Joel Beasley at 00:46:23) And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email: [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.