Building with Low-Code: A Product Leader’s Story

Introduction

Hi, everyone. Thank you very much. Thanks for the opportunity to speak.

Personal Background

Background, I come from... My name is Ohad. I come from business. I was in McKinsey. I was at ING leading product teams. I was in a startup, GM, kind of doing all of the introduction to new markets.

Entering the World of LLMs

Last year, I had a... Coinciding with the moment that LLMs are starting to get into our world, took it as an opportunity and said, you know what? I actually want to get my hands dirty. I want to build stuff.

I have an eight-month-old baby. I want to feel and see and make them go live. This LLM thing is super cool. I need to figure it out.

And I think over the last year, What I've managed to do for me was, one, immensely exciting. 1I actually managed to get things to become live without technical development knowledge, without syntax, without knowing how to write code. So I thought it's interesting for many people who are kind of in that phase.

So I, with that, have a company, Intent AI, trying to revolutionize core activities with AI-driven solutions. And let me kind of take a second and organize the thoughts.

Understanding Local Development

So what is local development, right? local development took the definition from perplexity, which is the new source of truth, as Google is dying, not dying, but you know, having their issues. So local development platform for application software with a graphical user interface to reduce traditional coding, right?

So kind of Building apps, building stuff that work, digital communication, digital language, experience, all of those lovely things that as a product manager at least I always was trying to bring to the world. But in a graphical user interface, it's a language that I can actually understand and communicate with, reducing the coding threshold.

In reality, what it is to have the low code means other people's code, right? It's like someone else has packaged stuff, made it manageable enough for people who don't want to understand all of the nuts and bolts underneath the surface so that they are able to actually do the stuff that they want to do and get it to go live.

The Intersection of Low-Code and AI

So why this? Why low-code and AI? Where is that those two things come together?

I think that, at least for me, when I was thinking about low-code, when I started touching these two things together, AI was really cool. I started GPTing, I wrote my prompts, and then it kind of stopped there. I didn't know where to place it or the copy-pasting from there to my Google Doc or whatever.

It was always very limited. If you use that and you place that in the Figma prototype, it's nice, it's cute. It's not the same effect as having someone who is able to interact with the things that you have put together.

So kind of what low-code does is it enables this really fast building and experimentation in a layer, in a level that you cannot do with JetGPT alone, right? Or Claude, or we can go deeper. 1For me, what that meant was real empowerment.

So I think as a whole, and I will mention that a bit later, As a product person, the challenge always was that you understand the requirement, you understand the needs, you talk to the dev team, and you tell them, OK, let's make this happen. And then there is like 17 layers of questions and topics and content and stuff to unpack, which is what's needed.

But my ability to go into those layers and to have that level of interaction that is helpful for them, but in a language that I speak, so not in is there a comma in the right place, Today, I am able to do that, right? Because I've actually built at a higher level, the level that I talk, and I'm able to speak to developers in a completely different level. And of course, faster insights.

So when I put all of that together and I put it in front of people, I'm starting to learn. I'm starting to see things happening. So it's overall really exciting, at least for me.

Exploring the Low-Code Landscape

And then I thought, I'll take a second to create what is my perspective on low-code, and it's a massive world, okay? The concept of low-code eventually, like any software that you can manage with a click to move is sort of a low-code. I took the pieces that connected to me in this period, in this context, right?

So I organized it, very basic, front-end, back-end. and integrations, front-end being what customers see, back-end, how the data is being organized, and external integration. Placed some tools.

I'm not marketing any of them, nor did I do a full market review. There is hundreds. But just to give a quick glimpse what I at least talk about when I say front-end low-code, so you have the Web Flows of the world, or Wix, which are very high level. It's kind of making pretty websites with a bit extra.

On the other edge, you have Bubble that does full end-to-end. So it gives you both the front and the back. You can manage your data within it, so it's important to understand what it is.

I use the one in the middle, WeWeb. I found them through one person. It's not like a massive thing. It's working wonderfully for me.

If anyone wants to work through and understand that a bit better, I can always share more. On the back-end side, it really gets more tricky. So you have things like Airtable, which is kind of light and pretty, but not really functionally complex in solving.

The two main ones that I've seen so far, Zano and Supabase, I thought it's important to give people a bit flavor of what that could be, what tools there are. On the other side, Vercel, for example, for infrastructure, if you want to speed up your entire process of throwing things in the air, but you need code connected into this story more. And on the other, external integration.

So if you think about building something that connects with a chat GPT, or OpenAI API, whatever that is, if you only want to leave those, to create those integrations, those touch points in a meaningful way, the Zapiers and the make.com, excellent. The downside, they don't solve. They don't create a customer-facing experience, nor do they manage the data itself.

So you need a Google Sheet or an Excel. And I'm an Excel guy, right? So I'm happy with Excel. That's where my home is.

But I've learned something new now, so it's fun. And I thought I'll take a second, move away from the theoretical, and show you a bit. I can continue?

Show of hands, which one? Show? Show. Show. Let's show. We love Sho. Yes. Okay. So, Sho.

Real-World Applications

So, this one was one of the first pieces that I've built, right, roughly by myself, which is pretty cool. So, I am working with a store in the U.S. called the Keratin Store.

It has a large range of products that they offer, and we basically created a tool that Of course, you can write whatever you want, but what it does is it takes that, it throws, it starts the whole API process to an LLM, but it starts collecting more information. It unpacks all of the attributes of what it is that the person that has wrote.

If I said I have benefits of I want nourishing and revitalizing, right? It starts capturing what attributes should it capture. It gives you how to continue the conversation and then effectively adjusts and recreate a website.

that is fully tailored to your need, right? So what is the question that you have asked and give you back an answer that is why are the things that I want to propose to you are relevant for you, right? So all of the world of filtering, think booking.com where you have a thousand filters, but in natural language for any website.

Doing that, you cannot do with JGPT. Just like you can try, it will suck, and then you will be stuck, basically. So you need to have that layer that takes that one layer, one level further.

Another one which is, I thought it would be nice to kind of quickly show, we've taken, I've been working with a company that has, they do calls, right? They do telemarketing and checking if those calls are appropriate and compliant, right?

I was able to, and this is like days to build, which is incredible, taking each of those calls, transcribing them, so using all of the APIs and the nice capabilities out there, and analyzing each of those calls against a long list of what do they want to know about each of those calls in smaller chunks, in smaller pieces. So that kind of enables you to play around and have much more control of what it is that you're asking the LLM. So you're no longer, I need to throw the whole shebang at the GPT and kind of hope to get the right pieces. You can actually organize it the way a software does.

but you can do it yourself, right? And then from there to dashboard, et cetera, and we're talking about really short time to market. Now you can see it's not enterprise grade, no doubt about it.

And I will mention that in a second, but the fact that it's not enterprise grade does not mean that it's not extremely valuable. It is valuable both in terms of the insights that I get from working with a customer as a product person. It is valuable sometimes for the customers themselves if they're not a big, complex organization and all they want is to solve their current problem and they cannot do that with ChatGPT type of solution.

The Good and Bad of Low-Code and AI

I thought I'll give a bit more what is the good and bad and how much are we on time? Good and bad. Okay.

So just to, or maybe one last piece, I do want to show you for a second, something like this, so that you will have, what do I mean with low code? Because that will make it very, very real. So you remember the page that we saw before? Great.

So you can see how kind of it is extremely visual and playful to be able to do something. So if I want to change something and I want all of these to, instead of being, I don't know, going like, if I click on this, and I want to set them differently, I want to place these boxes to be horizontal, or it could be either what's called CSS, kind of how pieces are organized, it could be how the workflows of interactions between those pieces run, and all of that in a way that at least for me was extremely kind of intuitive.

So if I want a text to be connected to something, I just click on this and it's connected to a value that I managed back in the back. I define how it needs to be organized, or I define the workflow that pulls it together and runs the next piece. So that gives me kind of lots and lots of control and playability.

And the same goes with Zano, which is sort of a big Excel table for me and that I can communicate front and back to words, right? So there's like a sort of a big table that in my world I can speak to because I understand Excel. And then I'm able to run an API connection that talks and brings the stuff from here back up to the front, to the user interface that I created.

Good, so a couple of more back to summarizing the good, the bad, and some things to be aware of, right? So first on the LLM side, so what is the value? One of the questions I always ask myself is what is this AI actually good for beyond the chatbot, right? I love the chatbot, it's wonderful. ChatGPT is magical. But it always feels like there is much more that I am not yet seeing enough out there kind of concretizing, becoming concrete in the world.

So I think two things that I saw in the low-code world that connects very nicely. So one is all of the, if you're trying to customize something, if you're trying to do something that is complex and it's beyond my skill set when it comes to development, LLMs actually step in very beautifully, sometimes fully integrated into these softwares, and you just click a button and say, I wanted to do this, this, and that. and you see if it works or doesn't because it's visual, right? So if it doesn't work, the thingy breaks and you see that it doesn't come. So you don't need to unpack all of the code underneath to do it. Also integrations, which also have high complexity usually, You can talk to an LLM and drive all of how to implement those. That makes it a lot more flexible and customizable.

On the other side, it does enable you to do, so the two things that we just showed, chatbot interaction. So if you have in mind a use case that is chatbot-y and you want to create it, today it is off the shelf, almost fully available, click and play, and you are able to play around with a solution. that will deliver the things that you want to do in a matter of days. And LLM is a code layer, which is my personal kind of... Where I see LLMs really as interesting is that at some point, the way data interacts with each other. Today, still, LLM is something that is mostly sitting outside. You send something to it, you get it back. I am not seeing enough cases in which it actually changes the data transformation process. What you see on one system appears on the other. One of the use cases that I showed you is when I get a question from a user or when I get a catalog of products, I'm able to extract from it a taxonomy of meaningful things about those products and then call that back. That changes what it is that I'm using within the code and kind of a new layer of data that I can play with and LLMs are kind of enabling that new capability.

Maybe last one, a bit about the limits. As a whole, when we're talking about low code, It is not perfect. It is, as we said, it's other people's code. It's only as far as another person has thought it through or has decided to put effort into solving something for it. So you have pre-built components that have their limitation. Integration, I mentioned, that's why we need sometimes LLM to help us solve them. They're always a complicated thing, so if you are able to stay within the environment of one solution, life is much easier. And then the last two is about scalability concerns and refactoring. So we just had a chat about it a moment ago. When you want to transition from this really cool, really quick MVP ready to show thing to a proper enterprise grade solution, There is a step. There is a refactoring. It's not going to last. It's not mature enough, at least today, to do that transition. At least that's what I've experienced so far. Maybe listeners who are coming from the local world will have a different and more mature perspective on this. That's okay. It's not the purpose necessarily. So as long as you know where to use the tool to give value, it's fine. And with that, be prepared for that future refactoring. It will happen and that's okay. That's part of most software processes anyway. And this way you have a software process that a person like me is able to have a software conversation in depth about things that up until now was always distant because we were speaking two different languages.

Getting Started with Low-Code

How to get started? Well, just start. It's easy. It's fun.

It's like my wife sometimes is pissed at me that I'm wasting time building stuff in my low code instead of doing stuff that I should. It's really fun. I'm really enjoying it. Purely the most fun, like, yeah.

It's like building in Legos. I click, click, click, click, click, and things just happen. It's amazing.

As for advice, I can tell you I took, so my beginning of the journey with this, I actually took, I hired someone who's very good at this to pair programming, to kind of sit next to me and tell me, you need to click on that button. No, no, no. Then you put it like this. it did help, it did require, because as I tried to do things that are more complex, I started getting stuck.

Today, you can use LLM to solve some of that, not everything, right? So having someone guiding you, that helps.

I mentioned splitting the layers. I think, and I mentioned that before, that I was talking front-end versus back-end. There is complexity in each of those and being able to control them makes life easier, gives you more flexibility.

That's why I think kind of WeWeb that I've talked about actually enables you also to extract all of the code as code. So when you get to the scalability challenge, you just give someone the code and they play with it.

And I mentioned later refactor, that's okay. Don't be worried about it. And yeah, I think enjoy the ride.

Conclusion and Q&A

And with that, I think I will finish. One minute, nice. Maybe that's my final take. for a question or two.

Does anybody have a question? You said just start, but where did you start? Where did I start? It's the first website I should start with if I want to start. Or not website.

So first of all, when I started, I had an idea of what I roughly want. And then I went to Upwork. Upwork, which is for hiring experts, and I said, actually, no, I went to GPT, and I told it, GPT, dear GPT, I want a local tool to solve this, this, and that. And then GPT said, you should use Adalo, which I've mentioned before. Fine, GPT said so.

I placed it on Upwork, and I said, dear Upwork people, who knows Adalo? And then one of the guys reached out to me and said, you're wrong. This is the wrong tool for your need. You should take this one. I'm like, I like. I like the guy. He's the one who taught me afterwards how to use the tool.

So if you ask for my advice, what worked for me, it's WeWeb and Zano. But I'm not marketing anything. And if you want to chat about your specific use case, maybe I have a different advice. Or a smarter person knows more about this. I'm more experienced in low-code.

Any other questions? Amazing. Good.

Finished reading?