Introduction

So I'm Guille, co-founder of Zunder.

We're basically an AI agency and we work with all types of different solutions.

We've been specializing in AI for about a year and a half now, recently specializing in custom software and voice AI agents for different companies.

Exploring AI's Potential

I thought relevant today come to talk about the possibilities of creating your own custom software with AI tools, AI tools that have made a lot of noise recently, but are very undiscovered.

And I think that if used properly, you can create very cool solutions, both for your clients if you're an agency and working with different companies, or internal tools for the company you're working, or internal tools for yourself even.

1So if this gets too technical, let me know if you have any question.

I might talk too much about these tools.

Key AI Tools for Software Development

But to start off, have any of you guys heard about N8n or Lovable, which are AI tools?

Yeah?

OK, most of you have.

Supabase?

One, okay, okay, some more people.

Okay, so those are the three main tools I wanna talk about today that I think if you're a person wanting to build your own software are the three main players nowadays to build something that's actually valuable and that's actually a proper software that you can build without needing developers.

Building a Software Framework

Understanding Frontend and Backend

So to start off, I want to start talking about how a software is usually divided into main areas.

You've got the front end and you've got the back end in the softwares.

And you have to understand what each tool of these does to make proper use of them.

Supabase: The Backend Foundation

I'm starting off with Superbase, what you see in screen.

Superbase is basically a database where you're going to be storing all the data you're going to be generating for your software.

You can see this as your own CRM.

You can see this as an

very powerful Google Sheets, but it's actually a database.

So it's going to enable you to have millions of records and data points, and it's not going to lose speed.

It's not going to break as Google Sheet might do when it has more than 50,000 rows.

And same with Airtable.

You probably have heard of Airtable.

It's good, but it's not as powerful as Superbase because Superbase, it's a no-code database.

So it's built for people that don't necessarily know how to code, but do want to build very scalable or very ambitious softwares.

So this is the first one that's going to be part of the backend.

N8n: Automation and Logic

Then we've got NADN.

NADN, you guys, half of you have already used it.

NADN is...

Sort of a new player in the automation world.

It's actually been out for five to six to seven years, I think, but it just recently started growing a lot about a year ago because they started putting a lot of focus on AI.

So, NADen, it's an open source software, meaning that it's a completely free tool if you know how to implement it.

You were asking about free tools before.

There's a lot of tutorials on how you can download NADen and put it on your own server, and it's gonna be free forever.

as compared to make, for example, that you're going to have to pay for an amount of operations, and then it's going to be completely free.

And you get the added value that's built basically for AI.

It has a lot of native modules that's going to let you do a lot of cool stuff on AI.

And then I then just raised 50 million about three, four months ago, so they are gonna keep growing and it looks like it's gonna be the most powerful software for building AI solutions.

Don't get me wrong, there's a lot of AI softwares and there's a lot of softwares that may let you do automations, but I think Aiden is the one that you can build the most amount of different stuff all within one same software.

You can build agents, you can build basic automations, you can basically integrate it with everything.

So N8n and Supabase are what's gonna be the backend of your software.

Importance of Database Structure

When you're building software, we'll later get into Lovable because I think it's still very interesting, but it's always the last part.

When you start to build a software, the first and more stable thing you wanna do is build a database.

That database is gonna, you're gonna really have to think about that database a lot.

That's, I think, an error a lot of people that don't code

make at the beginning when they are starting to build software where you start building something that you think is very cool either for a client or a software for your own and after a few months or after having let's say 10,000 data points it breaks everywhere because your database is not properly set up because you're not building your database in a logical solution and because let's say I have a

tattoo marketplace app, which is something I built and I'll show you later.

At the beginning, we started with three tables, right?

One for artists, one for tattoos, and one for users that are logging in in the app.

Nowadays, we have, I think, like 25 different tables because when you start developing, you're going to see that you're going to build a table just for relating one tattoo to one tattoo artist.

Then you're going to create a table just for the likes of different tattoos as likes or comments on an Instagram post, for example.

So I don't want to talk too much about the database structure, but it's something that if you want to build a proper software and you want to build something scalable, you do have to spend a couple of weeks building the database.

And that's very important because if you don't do that, you might as well have to break your entire app later on.

Integrating AI with Software Development

Utilizing AI for Database Creation

Something that's very cool, now that you guys were talking about different AI apps, I've been using Cloud for a long time.

If you guys have used Cloud, they released MCPs.

Is MCP, that term, anyone know what it is?

Okay, two, three, four people.

Okay, so MCP is something AI released a couple months ago, which is a method for any AI model to speak with all your softwares.

Basically, that AI model, that cloud, that GPT can take actions of looking for any kind of information in your softwares, taking actions.

You guys were saying about the integration with Gmail, for example, about the AI model.

Well, with Superbase, what I've been doing a lot for the past few months is connect my cloud to Superbase.

And I'm not the one building the database itself.

I'm just talking for a couple of hours with cloud saying, hey, I want to build this software.

What do you reckon are the tables I'm going to need to make this software scalable?

What data types am I going to want to write?

These are the features I want my software to have.

So take into account when you're building the table.

So that process starts with me talking back and forth with Claude, having the help of Claude to be constantly reiterating and adding or creating new tables in that back end.

Automating Processes with N8n

Once we have this, NADen is really what we're going to use later to populate the database.

In the case of one of the softwares we've built with NADen, for example, we have a custom software for our clients for voice agents, for AI voice agents.

What we do is, the voice agent is built on an ADN and one other software.

When the call finishes, that data is sent to an ADN, we clean that data, we analyze that data, and send it to Supabase.

So it's gonna be what's, an ADN is what's moving the pieces together, it's what's grabbing the data from one place, sending it to another.

So it's kind of the logic that makes everything happen, and that makes the software actually functional.

Lovable: AI in Code Generation

1And then we go into the third piece of the equation, which is lovable.

Lovable is, if you guys have not played around with it, it's basically an AI agent that generates code for you.

You can make very bad use of it, and it will create horrible softwares, horrible apps.

Even if you don't use it correctly, you won't be able to create a good landing page.

1But if you know how to talk with the software, if you know how to connect it to a database, give it the logic behind, you can create very, very powerful stuff.

So with Lovable, we've been using it for about five to six months, I think.

And over the time, you're of course gonna need to play around with it a lot if you wanna build something that's actually useful or something that you want to sell.

But if you start working with Superbase and then just connect your Lovable project to your Superbase table, it's already gonna have a structure that's gonna be a lot easier for that agent to later build upon that.

instead of just giving it an idea and let Lovable generate what the database it thinks.

So when we come to Lovable, this is the basic interface.

Something that they've added recently that I really like, and these are just tips and tricks that I use when I'm creating with Lovable.

Lovable has two modes when talking with it, one's chat and one's just code.

I've noticed that if you first make it aware of the code that's already in the database and ask him for solutions, let's say I want to implement a new feature, I'm first going to use the chat function to tell it to review the entire code and give me a plan to implement that new feature without breaking anything else.

because I've seen also a lot when playing around with Lovable, I create a new feature and it breaks down 15 other features and that messes up the progress of a couple of weeks or the credits because Lovable is also a very expensive tool if you don't know how to use it well.

I think with Lovable, once you've got the handle of the other two, you can create very cool stuff.

Challenges and Opportunities with Lovable

And again, the idea today was to show you guys three very different softwares we've built, both internal and for clients.

If you have any doubts about the software or you've been having bad results with Lovable, don't think it's because the tool is not capable.

It's because it's actually very hard work.

or you require some experience or some usage to get something good going on.

In the end, with Lovable, we are playing a game of removing developers from the equation.

It's not perfect, it's just started, and right out the top,

of our head, we can't expect to start speaking with an AI and replace a team of 10 developers, but we certainly can get there.

We can start coding things that are very interesting and then have other developers work upon that, or as lovable improves, just continue having good softwares.

Showcase of Developed Projects

Tattoo Marketplace Application

So examples of things that we've built.

This was a project that we started a long time ago, and it took us about two weeks to build.

We still need to add some features, but it's basically a tattoo marketplace where we've created the authentication where users come in, they can log in and create an account as a tattoo artist, as a user, or as a tattoo studio.

And the idea of the application is that you're gonna come

to explore designs, for example.

This is all connected to my Supabase table where all the mockup data is set up.

And you're going to be able to filter the different tattoo designs you're going to want to see.

And you can enter, book a tattoo, and then you get in contact with the tattoo artist itself.

You can see that the software itself doesn't look like an AI agent has coded it.

And you can try to break it.

I'm pretty sure you won't be able to break it now because we've been putting a lot of work into it.

But you can see that it's not just for internal, four minutes, okay.

Not only for internal use, but you can build things that are interesting.

You later combine that with something like Stripe,

which is a payment application, which we've already integrated here, and you've set up a complete marketplace.

Now what you need to do is marketing and sell it.

But the connection is there, the database is there, the payments are there, so you don't have to really be focusing on

raising an investment to build an app or spend six months to build a very basic MVP, you can spend two weeks building something like this, go to tattoo places, talk with them, say, hey, give it a go, see how it goes.

If you like it, you can be part of a project if we start building it, for example.

That really depends on how you want to sell it.

Client-Facing Applications

We've got another example, and this example is our client facing application that we've built for, this is an example of one of our clients, which we have prepared for both voice agents and video AI agents, which are two of the models we've been selling lately.

In this case, this client only has access to video agents.

What having this software enables us is to not just sell an automation to that client where we sell video agents.

It lets me give a full software to the client, and that goes hand in hand to being able to charge a monthly subscription to that client because I'm just not building the automation.

I'm giving them access to a full software where they're going to be able to see all those KPIs.

and they're gonna be able to go to their video content tab, see all the videos that my automation has created for them, where they can approve, deny, reject, or recommend different edits for those videos.

Once they click that, of one of the videos?

Yeah.

The one I generated with VO3 is not there.

This is the testing account, so we have one of the videos.

Let's see.

These are fully, this flow that we were trying out was basically we gave it a newsletter that talked about current events and we had an AI agent generate the script, then one generate the music, generate images, images to video, and then we put them together.

Now we're changing the flow to VO3 because we're liking the results a lot better.

It's a lot more expensive.

Every single one of the videos for the client, it costs about 24 to 26 euros.

but if they're willing to pay, we just build it, charge the subscription, give them the software, and that's really up to them.

It's in Spanish.

This is the last one, I haven't even seen it.

Ah, wait.

This... Okay, this was a trial the client made.

which is not current news, but just something the client made himself.

What secrets does this velociraptor's gaze hide?

I whisper as the story begins.

So the idea there is these videos are going to be generated every single day without having to lift a finger.

This was when we gave the client an option to put his own prompt.

He asked something about dinosaurs with Ghibli, and that was what was generated.

And what we're doing for this client is basically two Instagram accounts that are gonna be posting videos like this for current events every single day on Instagram and TikTok.

But this is the lovable app that enables that clients go there, approve and reject.

So it gives us a value that now we're talking with the client about the subscription between 500 and 1500 compared to a subscription of maybe 100, 200 if I had no software.

And the value perceived by the client is not as much as having an actual platform.

Basic Applications and Testing

And the last one was the simplest of them all.

It's a very basic landing page that we've been testing out for a Facebook ads campaign.

For you guys to see that we use lovable, not just for huge softwares or softwares that we're gonna use for clients, but for very, very basic stuff.

It also takes a lot of time instead of using Framer, Webflow, WordPress, something like that.

I got this page, I think in like three lovable prompts.

So that's also very valuable.

and you called it off?

This one, let's see if I have it actually.

Comparison of AI Tools

And by the way, not sure if you guys have played around with other builders.

I've tried Bold, I've tried Lovable, I've tried Verso.

In my opinion, Lovable is the best out of them all.

Design-wise, how it responds, the chat feature is very good.

How good it connects to Superbase and just generally the logic behind it.

I think that if you're going to start, this is probably the best one.

But up to you guys really to try which one you're more comfortable with.

Okay, a couple more than three prompts.

Yeah, a couple more than three.

But very basic.

Here I could scroll to the top, okay.

It felt like three, yeah.

If I scroll, you can see that it's not such a long conversation.

If I'm going to one of the others, I've probably spent days talking with this, these agents, but it won't stop scrolling.

it's been more than two months of work, whereas the other one is enough credits for one week maybe, at most two.

But it just keeps going up.

So yeah, that's the difference in the complexity.

Conclusion

Cool, so I think on my end, that's about it.

If you guys have any questions about any of these softwares or how to start your projects, I think these are the key tools.