Pioneering Music Education with John von Seggern

0:00 John: The people are actually very important and it could never be an AI only organization.

0:21 Brad: Welcome, everybody,

0:22 to another episode of the How Many CTOs Does It Take podcast.

0:25 I'm Brad Hefta-Gaub,

0:26 and our regular co-host Scott Porat is not here today,

0:29 but we have a special guest,

0:31 John Von Segerin, who is the founder,

0:35 CEO,

0:35 and CTO of Future Proof Music School,

0:38 John: Yeah,

0:38 Brad: I guess.

0:38 John: that's right.

0:39 I have to do a lot of the things, although I do have a partner who does a few of them, but yeah, I have to do the CTO CEO.

0:46 job.

0:46 Brad: Love it. Love it. Well, why don't you introduce yourself,

0:48 John,

0:49 maybe tell us a little bit about kind of your journey to this point of future proof music school,

0:53 and then we can talk a little bit more about what you're building and how you're building it.

0:57 John: I have been a musician all my life and I was a pro bass player for the first 10 years of my career but I've also always been really into technology and learning about and using the latest technologies.

1:12 And so then I became a computer musician,

1:17 electronic music producer at some point and sound designer. I've had a lot of different roles in the industry.

1:25 But finally,

1:27 in the last years,

1:29 I ended up working in education and specifically online education of electronic music producers.

1:35 That's the specific niche that I work in.

1:40 Within online education,

1:41 I've always been trying to adopt new technologies for learning.

1:44 I get really excited about things that can help you learn quicker and understand things easier.

1:49 And so in the last few years, that's all been about AI.

1:53 And

1:53 Brad: I love

1:53 John: so

1:53 Brad: it.

1:53 John: I've been developing various...

1:57 AI features and functions and bots that you can interact with to help you learn.

2:02 Brad: Love it. There you go. Well,

2:04 it usually only takes a couple minutes in the podcast before we start talking about AI. So we landed there pretty quickly.

2:12 Cool. Well, tell me a little bit about the process, obviously, with your background. You didn't have a classic sort of engineering background,

2:21 but obviously dealing with electronic music and dealing and now building this platform yourself.

2:29 What have you tell me a little bit about that process? I'm curious to hear how you how you've evolved and and how that's affected your ability to.

2:38 Build your enterprise.

2:40 John: Yeah, it's been a it's been.

2:43 quite a process of change in the last year especially so all my life pretty much I've been I've been a power user of technologies I have worked as a technical writer and a customer support manager at a software company and a knowledge-based writer and you know various roles like that I was never actually a developer I never actually wrote code about the furthest I went as sometimes I was

3:10 copying and pasting javascript out of things to make little scripts work but that's about as much as i could do and in my mind i always said god i wish i would have gotten into developing at some point but it seems like such a deep and vast subject i've already invested in being a musician and producer i can't really spend the time to learn to be a coder so anyway fast forward to more recently i when i had the idea to start future music school a couple years ago

3:36 years ago i had always thought my cto would be a friend of mine named steven and he is a developer down in florida he made a platform called midi which

3:51 Brad: Oh,

3:51 John: is

3:51 Brad: yeah.

3:51 John: uh for

3:53 online keyboard learning and he's a great developer and he i worked with him on his platform but my work was more like well hey what if it did this you know

4:03 Brad: Right.

4:03 John: and then he would implement it so and i would uh i had all the users so i would do testing and get user feedback and stuff for him but anyway i thought we would do this together and then when it came time i was okay steven i'm finally gonna leave my company you know and

4:18 and start this new school and he had had a baby and also decided that he needed to finally leverage his talents to make some actual money and so he got a really good job in cyber security which is great for him but um

4:34 Brad: Having kids will do that to you, make you feel like you need

4:37 John: no

4:37 Brad: to.

4:37 John: no doubt so he couldn't forge on with his uh low-paying startup life any longer at that point but anyway then he couldn't really be part of this

4:45 so this is about you know a couple years ago i wanted to integrate i know how to like use online platforms and design courses you know that stuff is pretty easy but i wanted to integrate ai into it that was the main thing i wanted to work on and so

5:04 At first, I thought, well,

5:06 you know, I could hire somebody to do a little bit of work on this or that,

5:09 but I realized like not only would that be expensive,

5:14 but the way these technologies are evolving,

5:17 it's not like we're just trying to build a chatbot or something and then that's it.

5:22 Like it's going to I could see this is going to be a continual process of evolution that's going to improve every month we'll be able to do things we couldn't do the month before.

5:29 So that means I'm going to have to develop it all because otherwise,

5:33 how could we keep changing it and iterating it, you know?

5:37 and

5:37 Brad: Yeah.

5:38 John: so at first i started working with uh no code platforms like i used zapier for a long time so

5:45 Brad: Yeah,

5:46 John: sure. i thought maybe i can use something like that and then i got into using nadn which was a little more powerful and you could make ai agents in that and i still use that some but what's really been powerful is since like six months or so ago

6:00 When cloud codes started to get really powerful,

6:03 then I started to switch over to use that directly.

6:07 And since then, I've built most everything with that or some other. I use other agentic coding platforms now,

6:14 Brad: Yeah.

6:14 John: too.

6:14 Mostly I'm using cloud code and I've taught myself how to architect software projects and.

6:26 you know make a detailed plan of what i'm trying to do and i've kind of learned the common obstacles you run into i've learned how to design a test suite and so i've become some kind of

6:37 Some kind of software development.

6:40 Good enough for my own uses anyway.

6:43 I've been able to accomplish the things that I wanted to accomplish.

6:46 Brad: Yeah,

6:47 that's awesome.

6:47 I think that to me, that's a pretty interesting example of what I am a believer that the technology is capable of doing.

7:00 I'm kind of curious,

7:01 would you like to say more about...

7:02 I'm curious more about what future-proof music schools platform,

7:08 what does it do? How do I use it? What kind of features does it have?

7:12 Tell me a little bit more about the actual product.

7:14 John: Basically what it does is we're trying to help people reach their musical goals as directly as possible.

7:22 So instead of having a fixed program of courses,

7:25 which is pretty much the case at all the schools I've worked at before,

7:28 you take some survey at the beginning and it recommends a program of courses for you that it thinks you need.

7:37 You can add more to it. You can talk to the AR as it gets to know you better than you can talk to it about what courses you might need to accomplish your, you know, your own ends.

7:49 And so we have an AI chatbot that works by text or voice. It's named Cadence.

7:56 and it is integrated with the curriculum as well as the students history on the platform and then we use that in various ways so that the simplest way is every unit you're on you might watch a video and learn something and then you can if you don't understand it the chatbot can explain it to you more or can explain how it relates to your music or how you might use that concept in your own music

8:21 Brad: That's

8:21 That sounds really,

8:22 really interesting. I'm kind of curious. So clearly there's a, you know, kind of the classic like learning module learning system,

8:30 you know, like the your

8:32 John: Yeah.

8:32 Brad: lessons and all that.

8:33 Is that are you using something white labeled something something off the shelf or have you also homegrown that component?

8:41 I'm kind of

8:41 John: No,

8:41 Brad: curious.

8:42 John: I'm using a platform for that. I'm quite...

8:45 Because I've had these different jobs,

8:47 I'm really familiar with that space

8:48 Brad: Mm

8:48 John: and what

8:49 Brad: -hmm.

8:49 John: platforms are available.

8:50 I'm using a platform called Learn Worlds,

8:52 which I really like. They have been really at the forefront of adopting new features and AI and otherwise into their platform,

8:59 and it's extremely flexible.

9:02 A lot of my Learn Worlds pages,

9:03 I have the thing in Learn Worlds, but I have something embedded from somewhere else, you

9:07 Brad: Mm

9:07 John: know, that

9:07 Brad: hmm.

9:07 John: I've read,

9:08 but Learn Worlds kind of makes it all hang together.

9:10 And also Learn Worlds keeps track of the subscriptions and the payment authorizations and all that kind of stuff.

9:15 Brad: Yeah,

9:15 cool.

9:16 And then obviously you had to integrate your agent with it, right?

9:19 So it's

9:20 John: Yeah.

9:20 Brad: got to be able to have like the context of.

9:23 the, you know, the courses, the coursework, the courses that the student has taken.

9:28 John: Exactly. Yeah.

9:29 Brad: Yeah.

9:29 So that's like an obviously interesting integration problem.

9:33 John: Yeah.

9:33 And so what's interesting about that is that working with these AIs, it's kind of like a plumbing problem in a way.

9:41 You have various things that come in.

9:43 On some level, it's quite simple, actually, because you have some things that come in and then you have a black box.

9:48 box that does something and then you have things that come out and if you're careful you can get it to give very precise outputs in the right format that you want but what's interesting is that you can't just load it with all the context in the world and hope for the best you know it actually works the best if it has the least context possible but just the relevant context that it needs at that moment

10:13 You know what I mean? So in

10:14 Brad: Sure.

10:15 John: some way, it's not a hard technical problem.

10:17 It's more of a logic problem of trying to think like, what should the agent know and when should it know it? You know what I mean?

10:25 Brad: Yeah,

10:26 that's super interesting.

10:26 I'm kind of curious.

10:27 I assume like, you know, when I wear my CTO hat,

10:30 I immediately like kind of go to, oh,

10:32 how would I solve this problem,

10:33 you know?

10:36 Obviously, like you're dabbling in essentially,

10:39 you know, kind of the rag space of trying to provide the LLM some context and stuff like that.

10:45 I also jumped to the idea of like.

10:48 implementing like i don't know if learning world has like an mcp server or something like that but that it strikes me that like if the agent was able if you were able to give the agent tools mcp style tools to say you know query against the the content you know the the learning content query against the student that might be a

11:09 So that's kind of the way my brain immediately jumps to now MCP servers are kind of kind of newish. So I don't know if they were really practical when you start working on this.

11:18 But I'm kind of curious is have you have you looked at using MCP? Do you use MCP servers in it? Do you have you looked at that as part of the pathway to it?

11:25 John: So first of all, I have two AIs. I have the public facing agent,

11:29 which helps the students.

11:30 And then, of course, we have another one that works with me and helps run the school.

11:34 Brad: Sure.

11:34 Sure.

11:34 John: And that one uses tons of MCP services. It's tapped into all of our services.

11:40 learn worlds did not have an mcp server so i i made one actually that was one of my software projects and it's not really i mean i can't say that i understand the exact details of how it works but basically you know you're just gathering together the api endpoints into a bundle and you're saying this is this is the mcp server i couldn't access the content of the pages though

12:02 because

12:02 Brad: Oh,

12:03 John: alarm

12:03 Brad: interesting.

12:03 John: rolls doesn't let you do that so this turned out to be more i've done all this stuff now but it turned out to be more complicated than i thought because initially i just thought oh it'll just read the page that you're on and then it'll know you

12:14 Brad: Right,

12:14 John: know it seems so easy but because

12:16 Brad: right.

12:16 John: it can't it can't pull it from the page so instead i had to make this whole thing where you

12:25 I've made a web scraper that

12:28 Brad: Mm hmm.

12:28 John: logs in as me this is the only solution I could find it logs in as me every week and scrapes my own website and makes a mirror database of the content of every page and then my agent actually searches that and

12:44 Brad: Yeah,

12:44 John: it just it just has like the text content of every lesson basically and

12:48 Brad: sure.

12:48 John: then it matches it it matches it just matches the URL you're on it's like oh you're on this page

12:52 it just doesn't must be this but that seems so easy but setting that up was such a pain in the butt

12:58 Brad: Totally. Well,

12:59 it's like I take your point earlier about like kind of working with these things, it's kind of like, you know, they're kind of designed around the notion of like input.

13:10 process output and so if you just kind of can glue it together correctly and so you you

13:14 John: Right.

13:15 Brad: had this problem like you needed to get the input of the content of all the learning modules there was no way to do that with the particular lms that you had and therefore like and they didn't have an mcp server and therefore you had to like come up with a solution so scrape it create a side database do that it's like that's that's definitely like a

13:35 to kind of figure out how to plug the things together to get the data that you wanted.

13:40 John: That's actually it actually brings up an important point because for a lot of online platforms and services, I think.

13:47 making it exposing it to AI agents is going to be a huge feature you know they're

13:52 Brad: Yeah, I mean, I have to assume that like hopefully the learning management systems out there are like, yes, we're going to figure out how to be, you know, AI enabled and would give, you know, would hopefully like hopefully learning world is going to build an MCP server.

14:08 John: putting their own AI features in it but they weren't anticipating that

14:12 That somebody like me would want to control it all like the thing that bugs me the most at the moment actually is like I would like to be able to take a bunch of my notes and then create the courses with my AI assistant.

14:24 But it doesn't have the endpoints to do that.

14:27 So I've been trying to, I've been, the last few days I've been trying to teach Claude how to use my,

14:34 the admin interface of our site through browser use,

14:39 Brad: Totally, yeah.

14:39 John: which is kind of slow and chunky, but it mostly works, but it gets discouraged and it keeps coming back and it says,

14:47 it would be so much easier if you just did.

14:50 I'm sure it would but it wouldn't be easier for me because then I would have to do it I would really prefer if you would do it

14:58 Brad: That is so funny to hear.

14:59 Yeah, you know, obviously,

15:00 like things are changing so fast in this space. And I know that one of the things that like the different models,

15:07 whether it be either cloud code or the cloud code application or codecs, all of them are sort of.

15:15 of building out this like how much can we do on your behalf right how much can we run the browser on your behalf and i think i just saw something yes you know recently like in the last couple days about a new a new feature in claude desktop

15:32 basically just like starts you know controlling your computer mode

15:36 John: Yeah.

15:36 Brad: and just like starts doing it um and i'm sure that you're not the only one that's run into that problem right like that's the that's part of the where that stuff is coming from is people like yourself being like hey i want to do this thing there's no there's no backend there's no cli there's no mcp server but you know it would be available through the through the browser you know so it's it's

15:57 It's definitely interesting how these things are moving.

16:01 You know, you obviously have a big background in doing electronic music,

16:06 using software tools to make music and to help people make music and to teach people how to make music.

16:14 Do you think that has given you an advantage over other?

16:21 say non non-technical entrepreneurs like if you were if you were a veterinarian and you were trying to like build a platform for veterinarians maybe your background in in the sort of deep software use that you already had and being a technical writer you think that's given you an advantage in figuring out how to use these tools or or what do you think about that

16:40 John: Yeah, yeah, sure.

16:42 I mean.

16:43 What that makes me think of in music, when you're putting together a piece of music in your music software, you have like the tracks of each part and then you're usually putting some effects and processing on each one

16:55 Brad: Yeah,

16:55 John: to make

16:56 Brad: totally.

16:56 John: them work together and sound cool and very often you're in a mindset where you're kind of thinking.

17:04 you're thinking through the path of the sound so you're thinking like oh it comes out here and then it goes through this thing and then you know what i mean and then it comes out as this that's why it sounds like that Yeah, and

17:13 Brad: yeah, totally,

17:13 John: that's very similar to working with those automation platforms where you

17:18 Brad: totally,

17:18 John: have some kind of thing that's tracing a path and going through branches and that kind of thing or

17:23 Brad: totally.

17:24 John: just working out the logic of how something's going to work you know

17:29 Brad: You know, one of the things that I think is out there in the zeitgeist, and I want to ask this in the spirit of like, just to be clear, like, I think it's really cool what you're doing. And I think that you're a good example of like a proof point of how AI coding has, you know, kind of democratized like building software. And

17:54 But I'm kind of curious,

17:55 do you have concern,

17:57 one of the things that's in the zeitgeist is, you know, developer-y type people,

18:01 you know, classically trained engineers saying,

18:04 oh,

18:05 these things, these vibe-coded things are all just going to be fragile and brittle.

18:10 They're not going to, they might have exposed security issues or they might have scalability flaws.

18:17 Do you worry about that?

18:19 John: Oh, yeah, I mean, I think those are very...

18:20 very valid reasons no doubt but um i have personally spent quite a bit of time well there's i have spent quite a bit of time number one trying to understand what i'm doing so i understand the different systems i'm using and how they work together and i'm not just like vibe coding i'm like making a detailed plan before i start if

18:43 Brad: Sure.

18:44 John: you're not doing that then

18:46 it's very easy to get into some slop mode where you don't really understand what's happening and everything is just like running like spaghetti you know the security issues are very real and i think my stuff is in good shape because i've done a lot of security audits of my own things and i have codecs and claude audit each other very

19:04 Brad: Yeah,

19:04 John: often um

19:05 Brad: sure.

19:05 That's a good technique.

19:06 John: and then i've seen i've found various checklists like here's 25 things you better make sure you do

19:13 ship your app i'm about to ship my first desktop app actually and i'm really paranoid about that because

19:21 once it goes on somebody's computer then they have a piece of my code right and then also my desktop app is using gemini live voice and

19:30 Brad: Mm hmm.

19:31 John: if somebody got a hold of my gemini key there's all these horror stories of people losing hundred thousand dollars in google cloud fees because they're gemini that's like the number one thing i'm worried about is your gemini key gets lost or hacked so

19:46 I have been super paranoid about those. I rotate my keys, you know, but it would be very, I think it's very easy to, yeah,

19:53 it's very easy if you don't know what you're doing to get into something where you've done something really risky and stupid and you don't realize it.

19:59 So I think that's very valid.

20:01 But there's ways if you're thoughtful,

20:03 you can learn you can learn the things.

20:05 Brad: Yeah.

20:06 Yeah.

20:07 I love it. That's a great point.

20:09 Like if you're thoughtful,

20:10 you can learn how to do it right.

20:11 John: I mean, you can have Claude teach you how to do it.

20:14 But I think also about the code. Another point that's good to think about is with the code being fragile or

20:22 mal malformed in some way i'm sure that the things i'm building are not the most elegantly constructed i don't even understand how they're constructed at some level but um the models are getting smarter as you go along you know and so already with some of the apps that i made from the beginning i loaded it

20:46 lately i've been using codecs more it seems like it has a little bit of an edge right now so i've loaded into codecs i'm like hey we made this over the last six months what if we i'd like you to analyze exactly what it does and let's figure out a new way to make it that would be more elegant and work better and not have all these crazy issues or whatever and as as the models get smarter it's easier to refactor things or rebuild them

21:12 So ultimately, ultimately, you'll just be able to be like, hey, make this so it doesn't have any problems.

21:17 Brad: Yeah.

21:17 John: You know what I mean?

21:18 Brad: It's funny you say that. I was listening to some other developers kind of talk about this problem and one of them was like saying,

21:28 and I think I agree with this perspective. They were just like,

21:31 look,

21:32 you know.

21:33 actually the code that these things are writing right now is is pretty good code now a year ago 18 months ago you

21:40 John: Yeah, yeah.

21:40 Brad: know some of it didn't look that good but now it's actually pretty good code and then you know the as they were talking about they were joking around about it they said one of the guys said like have you looked like how many times and i could say this as a as a cto who's been dropped into many organizations i'd show up and like do a sample of the code and you'd look at it you'd be like oh my god this is

21:59 This is garbage.

22:00 Well,

22:00 John: Yeah.

22:00 Brad: it was written by it was written by engineers. It wasn't written by AI.

22:04 So like human engineers wrote wrote junk code for decades,

22:10 you know. So it's it's it's a little bit of a I think it's a little bit of a potshot to say like,

22:16 oh,

22:17 the AI is going to write bad code.

22:18 Now,

22:19 in fact, AI writes pretty good code compared to.

22:23 The average.

22:25 Developer.

22:26 John: but you know you know what happens is that the ai will just if you're not if you don't understand what you're doing and you're just like hey we need to make sure this works it'll just reach for the most expedient short-term solution sometimes and then you end up with all these little things glued together or something whereas if you started from the beginning and you were like we're making this thing and it's going to work like this and have all these things then it can just architect the whole thing perfectly from the beginning

22:48 Brad: Yeah, I think that's a fair point.

22:49 It also,

22:50 I've also seen cases where it over engineers.

22:52 engineers things so under engineers and over engineers and the point

22:54 John: Yeah

22:54 Brad: that i think you're making is that as the orchestrator of the whole thing you have to kind of give it like you said the right context you have to kind of be clear about your guardrails be clear about what you know your big picture plan is and then you'll have a better shot at actually like staying on target

23:12 John: You said about the famous phrase,

23:16 the democratization of code,

23:19 which is the same thing that's the democratization of video production and the democratization of music writing and everything else.

23:26 But at least in code,

23:28 it's been super empowering for me.

23:30 But unless I got more education about what I'm doing or more training,

23:35 I don't think I could do really bigger.

23:37 bigger projects than what i'm doing like i made a dashboard for my students that's like within my wheelhouse but um

23:48 I couldn't make like a secure payment platform that thousands of people use every day.

23:53 Brad: Sure, totally.

23:54 John: You know what I mean?

23:55 So there is a limit to how far you can push this.

23:57 And also I read a lot of posts from people online about using these things and it's become pretty apparent to me like the people that started out that were already developers like my friend Steven.

24:13 are able to do 10 times more than I am.

24:17 oh because they already had that knowledge i can i envy the people that are or who seem to be setting up these pipelines you know where it's like my design agent does this and my coding agent builds it and my testing agent does this you know and i was like wow that sounds so cool i don't really quite have the time or the knowledge to build up

24:36 Brad: It's interesting the way you're framing that,

24:38 like someone who already has the experience, they probably are going faster. I've been saying that for a while.

24:42 I feel like I've kind of described it as like the K-shaped economy of agentic coding,

24:47 John: Mm

24:48 Brad: like the people

24:48 John: -hmm.

24:48 Brad: that were already strong engineers who lean into it are going to are going to be accelerating even faster. And the folks that the folks that were,

24:57 you know, didn't have that background are going to I think it's not just about engineering experience, although I think that's a big key part of it.

25:06 But I think maybe your story is an example of.

25:08 of how powerful curiosity is and how and so you you basically have approached this problem with you know an open mind and a curious mind and a and a and a problem solving mind right so you're just like I'm going to figure out how to do this and I'm going to and I'm going to work on it until I can figure it out and I think that's probably benefiting you really well and if even if you I mean I have seen in my own

25:35 immediate experience people with lots of engineering experience who simply didn't want to have a curious mind about it you know and that's not that's nothing to do with AI

25:47 As waves of new technology have come through the industry,

25:51 you know, I've been I've been writing code for more than 40 years.

25:55 I've been, you know, leading engineering teams for more than 35 years.

25:58 And how many times have we had new technologies that come along and we and.

26:05 Each of those waves of new technology experienced engineers in the old technology will sometimes resist the new technology.

26:13 The ones that are curious and embrace the new technology are the ones that you see this rapid acceleration in.

26:20 I think that's certainly the case.

26:23 A chat to coding and AI.

26:25 John: Yeah, definitely.

26:26 And what's really interesting too is that the basic skill or concept of like supervising the agents doing things for you, obviously you can apply that to so many areas.

26:38 It's not only about code.

26:40 And if you can...

26:43 get your mind around that and figure out how to do it's incredibly powerful i'm sure you understand exactly what i'm talking about but and then you look around at you know the people you know and it's like whoever you know if they can grasp this one thing they'll be then 10x you know more productive of whatever they're doing and so again that's why you follow into like wanting to evangelize your friends and stuff but

27:09 I guess the curiosity is a lot of it.

27:12 I don't know what else it is, but we can talk about democratization of coding,

27:17 but I don't feel like most people are going to be doing this. It's still pretty complicated to think about

27:25 Brad: Yeah.

27:25 John: how to set up all this stuff and use it.

27:27 It might be like it's easier to make an app using cloud than it is to write the app by hand, but it's still pretty complex.

27:35 If you want it to work right, you know, so

27:37 Brad: Yeah,

27:38 I think that's a point that's well made, that at least today,

27:43 it's still pretty complicated to get this stuff up and running.

27:47 Now there's services out there like Replit and Lovable and things like that that are really trying to lean into the,

27:53 you know, just describe it and get yourself an app.

27:58 I like your point. I want to kind of circle back to the point you just made about.

28:02 And managing the agents is a lot like, well,

28:05 I think I heard, I know this is my opinion, and so I might have just confirmation biased heard what I wanted to hear, but it sounded like you were saying that a part of the becoming effective using AI agents in particular is thinking about and being capable of.

28:24 actually thinking about managing these agents that are doing work on your behalf.

28:29 And that that applies not just to code, but that applies to any back office thing,

28:34 right?

28:35 John: Yeah,

28:35 Brad: Like you mentioned

28:35 John: yeah.

28:35 Brad: that you have agents that are helping you run the business.

28:39 John: Well,

28:39 you know, my initial thoughts since I've been working on this now for at least a year.

28:44 So the back end agent we call Nova.

28:49 Initially, I was building it in N 8 M and I had this brain and I just kept connecting more services to it, more MCPs, you know, and more different things.

28:59 that started to get really bogged down and after a point it didn't work it my my fantasy of it was if we just plug everything into it it'll be able to read all of our data and tell us what to do you

29:11 Brad: Right.

29:11 John: know what i mean but then it's like oh well this doesn't work and it's like it's got too much of this and you know there are so many problems with it so then when i then when i got clod code i just started doing that maybe in january

29:24 So then I started over with cloud code and I said,

29:27 okay, we're going to make a brain that controls everything and reads everything and helps us with everything.

29:32 And I just.

29:34 kind of designed it together with Claude.

29:36 Claude and I together we focused on what are we trying to do with this?

29:40 Like

29:41 Brad: Sure.

29:41 John: what could it conceivably do for the business that would be useful?

29:45 What would be the most useful?

29:47 And then one by one we've just added all these different capabilities to it.

29:51 And these days so many new things happen which I usually learn about I was just working on this earlier because it had a bug but

29:59 but i read about things on x and i bookmark them and then i have this process for my agent and every day it reads all my new x bookmarks and evaluates each one and if like if one of them is like here's a new skill for claude code it'll just go find it and install it for itself

30:16 Brad: Oh,

30:16 John: and like

30:17 Brad: nice.

30:17 John: so

30:17 Brad: I love it.

30:18 John: it's it's like things are moving so fast like i don't have time even to

30:23 fully understand these things myself you know it's like wow this is a great thing that we could probably use while you go figure it out and if we can use it then let me know I it's

30:32 Brad: love it. That's awesome.

30:33 That's awesome.

30:35 I think you mentioned earlier you have a couple other people on the team. Do you want to say any more about sort of what is the shape of future-proof music school?

30:43 How are you other than Nova and yourself? And

30:46 John: kind

30:47 Brad: there's

30:47 John: of a hype

30:47 Brad: no wrong answers to this, right?

30:49 Like

30:49 John: Well,

30:49 Brad: the yeah

30:49 John: it's kind of a hybrid organization because the people are actually very important and it could never be an AI only organization because we're teaching young artists how to be artists.

31:01 And so I believe that they need to learn that from older,

31:06 more experienced artists.

31:07 The AI more functions teaching the more foundation concepts. It's probably the most useful for that.

31:15 And as people get more advanced and they need to study with the human teachers more, we have a lot of live Zoom workshops every week and the students have one-on-one mentor sessions with the teachers.

31:26 The AI supports the teachers by,

31:29 for example, when they're going to have a one-on-one meeting,

31:32 the AI reads all the student records and emails the teacher and says, so-and-so is having a hard time with sound design,

31:37 why don't you try to help them with

31:38 Brad: that Oh,

31:38 John: or something

31:38 Brad: that's awesome.

31:39 John: like that.

31:40 But anyway,

31:41 yeah, to answer your question,

31:43 so I have a co-founder who was one of my top teachers at my law school and is a well-known dubstep artist.

31:50 So he's the director of marketing and he runs the social media and he's kind of the community manager and he's the...

32:00 he's designed quite a few of the courses as well we've got a few other teachers that just work on contract like just whenever we need teachers or doing workshops or whatever one of the things I've tried to do actually more to the point of this podcast like my idea all along like I had a number of people I know a lot of people in this space because I've had these different jobs so I have however many teachers I need I can find teachers that I've worked with before

32:25 but i kept thinking if i could get one of my people to be as into ai as i am then they would be able to do 10x more work and i have actually succeeded with my partner so

32:38 Brad: I love it.

32:39 John: he's a dev step and he's kind of like kind of like me he's a dev step producer and he's really good with music software and music production but he never did any coding or anything like that less than me

32:50 And in the last few months, I got him really into cloud code and now I think he actually just got a side gig.

32:57 making an app for some other company he just

33:00 Brad: So there you go.

33:01 Love it.

33:02 John: told me that a couple days ago i don't know what it is but

33:04 Brad: Yeah. So there you go. You're sort of spreading the enthusiasm.

33:07 John: yeah and so he is able to do things like um the reason i wanted him to do that in particular is because we're using mcp server to access meta ads and

33:17 Brad: Yeah.

33:17 Oh, yeah.

33:18 John: so you can like you can get insane levels of analysis with that you know or you forgot to switch that one

33:24 That one little thing around

33:25 Brad: Right.

33:26 John: campaign,

33:26 is it working?

33:27 You know that.

33:28 Oh, my God, that's such a that's such a relief from trying to access meta and work with everything.

33:33 Brad: That's awesome.

33:34 I love it. What advice would you give a non-technical or like yourself,

33:42 technical enough,

33:43 but not a programmer about starting their own venture and using

33:48 the current set of AI tools,

33:50 what advice would you give them about how to, you know, mistakes not to make or?

33:56 John: Actually, there's just one basic thing to know that you could remember,

33:59 I think would be helpful for everyone.

34:01 When I was starting to do this last year, you were just kind of going back and forth with the AI and asking your question and getting a response,

34:09 but the AIs have gotten smarter.

34:12 You're going to get the best results if you just tell the AI from the beginning what your goal is.

34:18 and as big as possible you know so you could if you're trying to start a donut shop you know you could say hey claude i'm starting a donut shop my goal is to have a successful thriving profitable donut shop

34:29 Brad: Yeah.

34:29 John: and then and then you just what the easiest way to do this is you just tell the ai ask me you know 100 questions about my donut shop so that you understand everything about it and then

34:42 and we'll make plans after that and it's really easy to proceed if you just have the AI ask you questions and then it dredges all the info out of your mind and

34:51 Brad: Yeah.

34:51 John: I do that on all the projects when I'm doing that's kind of a standard way to proceed but when you're doing a software project you have the AI ask you a bunch of questions about if there's any ambiguities or you don't understand anything about our plan you know I want you to ask me and you go through and you can really hone down

35:07 that way so if you just focus on that the AI can explain to you how it could be the most useful you know because it's going to be different if you're if you're having a donut shop you could say okay we have a donut shop how can you help me with the donut shop and then it might say oh well what tasks do you have to do every day oh well I have to

35:27 you know probably marketing the donut shop is where the AI could be the most useful for a donut shop and so you could say oh I have to book ads on these things and I have to have special deals on donuts you know and maybe it would then say oh I could track the deals and see which deal works the best or something you

35:43 Brad: Yeah,

35:44 John: know but

35:44 Brad: totally.

35:44 John: but the AI can teach you itself and then you can just learn one step at a time and if you don't understand something just ask it to explain it again

35:52 Brad: I love it. The AI can teach you.

35:54 If you don't understand it, just ask it to explain it to you.

35:57 That sounds like a great set of advice for 100.

36:00 John: Yeah, let's even remember it that way.

36:02 Very cool.

36:03 Brad: Awesome. Is there anything else that you think that.

36:06 the listeners of the pod should know about future-proof music school or other other um or more about yourself or anything else about uh the future the future's so bright yeah

36:16 John: The future is so bright afterwards sheds advice.

36:19 My advice is mostly about music.

36:21 Do you have any questions about that?

36:23 Brad: tell me i actually yeah tell me tell me what do you it's so interesting i thought

36:29 When I first heard about Future Proof before I went to the website,

36:33 I was like,

36:33 oh,

36:33 is this going to be like an AI music generating software?

36:37 And it's clearly not that.

36:38 I'm curious,

36:39 what do you think about the future of AI generating music?

36:49 John: I mean, there's a lot that we could go into there,

36:53 but I suppose on the

36:56 broadest level i think people will use ai to make music in various ways for professional musicians and producers i don't think that prompting songs is really that fun or productive there's a lot because they don't you don't have enough control over it and

37:14 Brad: Yeah

37:15 John: also it's not really it's not really fun

37:19 It's not because you don't get into that flow state like you do like when you're playing an instrument because you're literally thinking of words.

37:28 it's like a you know what I mean it's like a left brain if you're into that kind of thinking it's a left brain function where you're trying to spell out something verbally instead of just like going with the flow and feeling it I don't really enjoy it on that level but I do think I mean on the other if you want to look at it another way people have been using more and more advanced technologies to make music for the last you know

37:52 five decades and so I'm sure people are going to use AI to they already are to make loops and parts and vocals and whatever and some people will like it and some people won't for the school

38:06 We have had to negotiate that because there's a lot of musicians that hate AI music,

38:12 first of all.

38:13 Brad: Hmm.

38:13 John: And so we don't want to get, we're just starting a business. We don't want to get caught up in this whole firestorm of who hates AI music. And on a more practical level,

38:22 the students,

38:24 the students that we have are really, they're really, really into learning about all the complexities of producing music.

38:31 You know, they're the kind of people that enjoy learning about that.

38:33 about that the

38:33 Brad: Sure.

38:33 John: same people enjoy learning about the complexities of computer code i'm sure so

38:37 Brad: Yeah.

38:37 John: um they don't really want to use ai generators they're not really that interested in it i tried having some workshops on it and nobody was that interested in it lately i started having a workshop that's been more popular that's uh vibe coding for music producers

38:55 and so that that was more interesting we mostly been we learned how to make some audio effects and

39:01 Brad: Yeah,

39:01 John: we

39:01 Brad: totally.

39:02 John: learned how to make some like more marketing oriented things but um it's another another interesting thing about the AI coding is that you can potentially make your own little personal tools that nobody else would ever want or have or care about you know you're not it's not something you're making to sell to somebody it just does something for you

39:19 Brad: Totally. I can imagine a world where somebody who wants to create a particular audio effect or something like that, but doesn't have the background in DSP engineering,

39:30 which is a complicated background,

39:32 but has an idea of what they want this effect to be.

39:39 could totally use AI.

39:41 John: I mean, that's actually not that hard.

39:43 I've done that myself,

39:44 but the coding of the effects is easy now because that's a pretty much a standard thing.

39:50 But the thinking of what would be a cool sounding effect and how it would work like sonically on a sound level.

40:00 that's a whole nother area of of technical knowledge actually

40:03 Brad: Yeah, totally.

40:03 John: you know and like my partner has i understand this stuff pretty well myself but my i would say my partner knows but better than me and so he's been able to make some cool plugins because he understands kind of on a technical level what he wants the sound to do you

40:20 Brad: Right.

40:20 John: know

40:21 Brad: I am super inspired by your story. I actually think that like.

40:25 it's a great example of someone with an idea and a passion to build it and you know you just found the latest tools and you've run with it and I think that's that's pretty cool I am that's the part of AI that I'm bullish about that

40:41 John: Yeah,

40:42 Brad: entrepreneurs

40:42 John: yeah

40:43 Brad: like yourself will feel and find it so liberating I think that was what you called it you called it liberating and I think that's that's very compelling

40:52 John: Yeah, let's look on the bright side.

40:57 I mean,

40:58 actually,

40:59 you know, I feel like there's all these big issues swimming around AI,

41:04 you know, from the practical issues to the philosophical issues.

41:08 And but the fact is, it's it's happening.

41:14 You know, nobody's going to stop it.

41:18 and it's like it's like electricity or something nobody's going to un-invent it and it's going to do some good things and some bad things and none of us is in a position to really

41:33 like the bad big things that can happen like yes it might yes there is some chance that skynet could happen and we could all die there might be a very small chance but i don't think we could completely rule it out

41:46 but I can't do anything about that I'm not going to worry about that like what you know what I mean what sense does that make I'm not and and a lot of the other things like what I can do is I can help AI help people learn about music I

42:00 Brad: Yeah.

42:00 John: can help people learn about music faster by using AI that's what I can do that seems positive to me and that's what I'm going to do so

42:06 Brad: Right.

42:07 That's obviously your mission,

42:08 right?

42:08 You care about music. You care about teaching people music.

42:11 And more power to you, basically.

42:14 I think that's the way to sum it up, John,

42:16 I think.

42:18 No one's going to uninvent it.

42:20 And so for

42:21 John: Oh,

42:21 Brad: those of

42:21 John: yeah.

42:21 Brad: us with a passion to do good in the world and to try to help people, you know, grow and learn or achieve goals,

42:29 let's let's use this.

42:33 This powerful tool to help them achieve their goals and

42:35 John: You know, actually, now that you said that, I would add one more thing, which is just that sometimes we see these stories like big tech is going to use AI to create, you know, the George Orwell's surveillance world or something, you know, and definitely you could see, yeah, people,

42:53 bad people could use AI to do bad things.

42:59 But I think that makes it all the more important.

43:02 If you're a person that wants to make the world a better place, I think you should be.

43:07 I hope they will be embracing this technology and trying to do good things with it.

43:11 Because if people that want to make the world a better place just avoid AI,

43:16 I think it'll just be left to the bad actors.

43:19 That's what I would say.

43:20 Brad: There you go.

43:21 I love it. Well,

43:22 it's great talking to you, John.

43:23 Thanks a lot for coming on.

43:25 John: Great conversation.

43:25 Brad: And everyone else listening,

43:29 we'll see you next week on another episode of the How Many CTOs Does It.

43:32 Does it take podcasting? Remember to smash that subscribe button, rate and review.

43:36 All right, John,

43:37 nice meeting you.

43:38 John: Okay,

43:38 have a good day.

43:39 Later.

43:40 Brad: This episode of the How Many CTOs Does It Take podcast was produced and edited by Saffron Heptacow.

43:46 Our theme song was AI generated by Brad.

43:50 Please follow us on all the social media platforms at HowManyCTOsPod.

43:55 You can find us on YouTube,

43:57 LinkedIn,

43:58 Instagram,

43:58 Facebook.

43:59 Look for Ed's X,

44:00 TikTok,

44:00 and Patreon,

44:01 or at our website at HowManyCTOsPod.com. And remember to rate, review,

44:07 and subscribe on the podcast platform of your choice.

44:10 Thanks for listening.