AI + No-Code = Learning Revolution | Joshua Wöhle on Mindstone

Hey everyone, this week I'm talking to Joshua Vola who is the CEO of Mindstone, which is a platform that teaches non technical people how to use AI in their day to day lives.

Then two and a half years ago ChatGPT happened and the world was no longer the same. And it ran at one of our events and it did way better than I even thought the SaaS solution would have been. You end up with maybe giving two, three sentences of an explanation of what you want changed but. But what actually gets output is a very kind of developed prompt that gets put into cursor. This technology is by far more powerful than anything I've definitely seen and it's moving at a speed that is very hard to keep track of.

Hey everyone, this week I'm talking to Joshua Vola who is the CEO of Mindstone, which is a platform that teaches non technical people how to use AI in their day to day lives. Now Joshua is one of my favorite people on LinkedIn to follow and I've learned tons about him about how to use different AI tools. So I'm really excited to have him on the series this week.

Hi, I'm Joshua Wola. I am a software engineer by background. I was born in the Netherlands at about 13 years there, 10 years Switzerland, 15, 16 years London. Now built a few companies before. The last company I built was a company called super awesome that became one of the biggest kids technology companies in the world. It was acquired by Epic Games, the creators of Fortnite and it was building a whole bunch of technology to keep kids safe online. So we built the tech that went into the apps of Disney, Lego, Hasbro, Mattel, all of the biggest kids brands in the world and then started Mindstone. Mostly because I always loved learning, hated education and thought there might be a better way to help people learn. And then two and a half years ago ChatGPT happened and the world was no longer the same and we ended up going down the rabbit hole and basically focusing entirely on how to use this technology and everything that comes with it, specifically for non technical people because all the big organizations are working on how developers can use these tools and I'm a bit scared of the world that comes out when non technical people have no idea how to.

Amazing. Now we're going to talk about something that you've built using AI tools yourself, aren't we? And I guess the conversation started here. I saw you posting on LinkedIn that you were just about to spend $15,000, something like this on a piece of SaaS and then realized you could build it in an afternoon. What was the problem you were trying to solve?

Yeah, so this was really one of the aha moments for myself. I started. So we run the biggest practical AI community in the world, and we run live events all over the. And at some point, someone was asking me, wouldn't it be great if we could start to make connections in this community? Because there are so many people that really would want to meet each other, but there's just too many. You can never know who you actually should be meeting. So I started looking for something like this, and I came across a service that did exactly that. The idea was you basically upload a list of people and they end up reaching out to all those people, figuring out who they are, what they want to do, who they're looking to meet, then create the matches. And then you would end up with this service where for each event, we could, ahead of time, tell people you should be meeting these three people that are coming to the event and they'll be in the same corner type of thing. It was really, really cool. I jump on a call with the founder. He walked me through the product. The product does about 85, 90% of what I really wanted it to do, but they were still a nimble startup. And so they. Were they on the call, actually agreed. Yep, we're gonna. We're gonna get this to 100%. We're gonna make it work for you. And this was, this was a Thursday. It was the last call of the day for me. I had dinner scheduled with my wife. After we got really excited, we said, okay, on Monday we're gonna jump on a call again. And they would have gotten it to like 90% so that on Tuesday we could actually do a trial and then get it to 100% after. So I got really excited, started talking to my wife about it at dinner, actually, because it was part of the day.

Yeah.

And then halfway through explaining what this was going to do, I literally. I stopped talking. And there's something that. That drops in my head. It's like, I. I actually think I can build this. And to be fair, the rest of the meal, I was just thinking about bait shit, real, like, this is not actually going to happen. And then the next morning at 8:00am, I try it. And within, within an hour, I've got 85% of the thing working. And then another hour later, the entire thing was working A to Z. Amazing. Yeah. And then I just finished it up over the weekend. And on the Tuesday, it actually was live in production. And it ran at one of our events and it did way better than I even thought. The SaaS solution would have been fantastic.

I mean, it's amazing how fast you can get these things out into the wild. Now, can we take a look?

Yeah, absolutely. So everything is hosted on replit, so it's not the fastest, but you can, you can see, you just have like some of the meetups that have been put here. You click on. Well, you can create an event very simply. There's nothing much to that, but then you get this attendee interface. You very simply can just upload a CSV. The CSV populates the attendees and you can see different statuses for each person. So imported invite sent. Everything is done by email. That's the other thing. It's basically just sending a whole bunch of emails, getting people to opt in. And once they opt in, they go through. I won't be able to demo the interface itself that goes through. I'll, I'll show the rest in a second. But the what, basically what happens is the individual gets an email, say, do you want to go do this? They're then directed to a chat interface, which I don't think this test one works here. Let's see. The test does work. Okay, there we go. So basically what the, what the chat interface does is it starts asking you a bunch of questions about who you are and what you're looking for. At the end of that chat, it then extracts the data, it creates a profile and the interests that you have, who you're actually looking to meet. Then it starts rating you against everyone else that is going to that event, and it starts creating these different bits here. And then you can see the status of each of these. Basically what happens when you get a match, you get an email with some information on the match. You then say, yes, that is a person I want to meet. If both people opt in and say, yes, that is, we want to meet. There's another email that gets triggered with both of them in copy saying, hey, you're both going to the same event. You both seem to want to meet. Here you go, now I'll leave it with you. And here you have a little bit of data on the funnel, kind of where, where people are going through.

And so presumably that matching is. Is that done in a like really fuzzy AI way, or is there sort of effective like bits of structured data you're pulling out of it that you, you knew were going to make good matches?

No. So the other beautiful thing is I am not doing any particular matching. I'M literally throwing the entire thing into ChatGPT. Basically what happens is I take a profile and I then run it through a loop where I put another profile and say, match these two from 0 to 100 and if these two people should be meeting on a scale from 0 to 190 or more, then I consider it a match and otherwise not.

And then you then email everyone a list of their matches and say, hey, these are the people you should meet up with.

It's one by one. So literally, for every match that happens, you get an email, say, hey, do you want this? Yes or no? If yes, you then wait until the other one has opted in as well and then you send another email to introduce them to each other.

Nice. That's fantastic. Tell me a little bit more about how you built that. I mean, you. You've been in the AI space for quite a while. You've got an engineering background. I'm imagining your, you're just. Your process for building things is. Is relatively developed.

It definitely is, and it's getting, it's getting easier and easier. But I still have an edge for the fact that I have a bit engineering background. Like the. I almost always now start the process in replit. Replit is kind of my tool of choice to go from zero to something. It. It gets it further and further. I must say, every week they're getting a bit better and they're, they're able to push it further along. But what I really like is they have an integration with Cursor as well.

Yeah.

So you can just. Basically, when Replit gets stuck, I can just launch an SSH tunnel into Cursor, which gives me my own development environment. And then I use Cursor to basically get unstuck from wherever a replit got stuck. And then when I, when I'm past that bit, depending on the type of functionality I'm trying to add, I either will go back to Replit, like if it's database intensive and stuff, then Replit tends to be where I start. If it's purely UI ux, I will probably just do this in Cursor. When I say I do this in cursor, to be very clear, I don't ever edit a single line of code in cursor. I am using the cursor agent and basically I'm just going back and forth with the agent there to edit the code itself. Now I've even gotten to the point where I'm using Voice to text. I'm just using Super Whisperer to get into Cursor. Which then does the edits, which streams back to Replit.

Are you doing any structuring of your prompts? You're asking Claude or ChatGPT to write better prompts for you before you stick them into Cursor and relet?

No. So this is, I guess, the bit that we have an edge on, or I have an edge on as well, is that I literally teach how to build good prompts. So like I, I do, I do know how to build a good prompt. I've now gotten it to the point where. So Super Whisper has a. Their entire flow is they do speech to text, but then you can. It runs through another LLM before actually giving you the final text that it spits out. And so I've actually, I have a mode where when I'm using Super Whisper in Cursor, the LLM process that takes my text to speech, transforms whatever I want, and basically tops and tails it in a way that it transforms it into a prompt that Cursor really understands. So you end up with maybe giving two, three sentences of an explanation of what you want changed. But what actually gets output is a very kind of developed prompt that gets put into cursor and that then gets executed on.

Got it. And is that that kind of transformation that you're doing in Super Whisper and I guess you could be doing with text if you wanted to. Is that a single kind of transformation, whether you're writing a new feature or debugging or if you've got some slightly different use cases as well?

Even within Cursor, I've got different use cases. Indeed. Yeah. So you can switch through the different use cases in Super Whisper and then depending on the case you go through, you have different ones. The other thing is you can use Super Whisper depending on the app that you're in. So super whisper in REPL. For me is different to Super Whisper in Cursor, which is different to Super Whisper in my email, for example.

Okay. And for people who haven't dived into Super Whisper yet, how would you kind of suggest they get going? Because I've loaded it up. I've seen you've got these different modes. You can define your own modes for using different applications. What else should a kind of a novice know to get the most out of it?

So as a novice, they've done it pretty well where they set up, I think, three or four modes to begin with, one being email and one being what they call Ultra. I think Ultra is basically speech to text with some corrections on like grammar and punctuation. And stuff. I would, I would say initially just look at how much better these models have become. I used to, I used to hate speech to text and it's only about a month, month and a half ago that now I, now I feel genuinely less productive when I'm in an environment where I can't use speech to text because too many people are around me. I basically, I feel like I lost 20, 30% productivity. And so they just getting to the point where you use them to understand where the models are at now. That's kind of step one, just use it. And then step two is trying to. I think email is probably for most people, the thing that will be the most transformative because the Super Whisperer also has the possibility of having the context of whichever app you're in. Which means when I'm looking at an email draft, so long as the rest of the thread is in that same window, which often it is, Super Whisper has that entire thread. So when I literally just start talking about how I want to answer that email, I then have a transformation at the end which takes what I said, the context of the entire thread and then structures an email in a way that I can literally just hit send. All our emails by the way, were text to speech. I didn't write a single line when we were going back and forth to organize this.

Yeah, amazing. I want to come back to that point. Just that the logistics you're saying now there's a lot of environments where you can't use, you know, voice to text and you feel less productive. Does that mean that you're moving away from open plan offices or working from home more are there. They're kind of like real world implications to that.

There really are. Yeah. I, I had this day before yesterday. Yeah, day before yesterday I was at a WeWork and I ended up having to put myself in a, in a real corner because I, I didn't feel I wanted to trade off the additional productivity. But I did get a few weird stares where people were like glancing over and I did have to think, wait, am I saying confidential stuff in this email that's about to go out or not that we're fairly small compared to most things, so it, that wasn't a high consideration, but it definitely became a little bit awkward. I think that is likely going to disappear. There's this whole thing about was it bone conducting audio stuff? So hopefully this will be a thing of the past very soon. But right now AirPods are still the. The way that I do most of this stuff.

Amazing. We've spoken quite a bit about Super Whisper. Are there any other tips or tools you'd really recommend for people who are just starting to code their first few things in these AI tools?

So in relation to coding, I really would just start with those two. There's a whole bunch. There's Vercel, there's Lovable, there's Bolt New. Yeah, they're all good. Actually in most cases when you're not an engineer, Lovable tends to be easier to start with. I think the same is true for Bolt New. I went with Replit because Replit is the most capable of them and it also handles the infrastructure. So you basically you have a one click deploy mechanism that goes there. The biggest thing, even when you're building, I'd say using ChatGPT or Claude to just have them spec what you actually want to build, because especially when you don't have an engineering mindset or a product mindset, actually clarifying your thoughts in a way that makes sense and making sure that what you're telling the AI to build is actually the thing that you have in your head is probably the part where most people end up messing up. If you're extremely clear, the AI can execute quite a lot. But if your thoughts are a bit all over the place and they contradict themselves, then the AI doesn't quite know what to do with it and ends up making some decision that might not be the one that you actually wanted.

Yeah, yeah, I've certainly found that the case when I've been using them myself. Now look, we've talked a lot about coding and things. I feel like we haven't talked as much as we should have done about Mindstone because I know you're just doing such great work here. Tell us a little bit more about what you're doing.

So we do AI training or AI adoption for non technical people. We work with some of the biggest organizations in the world like Hyatt Corporation, the Hotel Corporation, Pearson, ebsco, Lufthansa, a whole bunch. And we help people across the organization understand how they can use these tools in their daily jobs. We talked a lot about coding here, but actually most of the things that we specialize in are all the non coding bits. How do you help people brainstorm using this technology? How do you help them build better presentations? How do you help them give feedback in a way that will be constructive and well received? On the other end, how do you build better strategies, vision documents, all of those things that really most people forget. And I think this is something I really care about, which is that this technology is by far more powerful than anything I've definitely seen and it's moving at a speed that is very hard to keep track of. But we have two potential futures that we are running towards here. One is a world where this technology continues growing but only technical people understand how to use it. And the problem is you then end up in scenarios where companies all over the world end up leaning into automation. And it is much easier to automate someone that doesn't know how to use AI than it is to automate someone that does in terms of their potential output. So in one world we end up automating a lot. In the other, when all non technical people understand how to use this technology, they are augmented. They can do two, three, four, five times as much and we all end up reaping much more of the benefits that come with the tech. And so if we forget to help people who are not technical understand how to use it, we end up in a bit in a dangerous kind of potentially dystopian future here where really we have to deal with the consequences.

I couldn't agree more. It's the rate at which things are changing is both exciting, but it's also scary. And I think the more people that learn those skills and those tools, then as you say, there's value all round, not just for individuals and companies, but for society as a whole.

Yeah, exactly.

Well, look, we have to wrap up there. Joshua, you've been very generous with your time. Thank you so much. We're going to stick your social media links and those to Mindstone in the show notes. So if people want to follow up or learn a bit more, they know where to find you.

Perfect. Thank you very much for having me.

Been a pleasure.