Welcome to this episode of WaveLength. Who is not talking about AI today. If you are not working on it today, it’s surely on your roadmap. And if it’s not on your roadmap, it’s surely on your mind. So it was for us in WaveMaker too. The possibilities of using AI to make imagination a reality and what it means for low-code have been on our minds. But they are endless. So we put our heads together and distilled it down to a few challenges that we think are stopping low-code from realising its full potential.
One of them is the effort it takes to convert an idea or intent into working code for enterprise software. The intent is best expressed not as words but as pictures. We started by finding a way to convert Figma (just because it’s popular) files into a working app in a single click (complete with mock API signatures and data). It’s not perfect but it resonates with early-bird users and has shown us that we are on the right path. Our goal is to give users pixel-perfect UI with the continued acceleration of low code and a strong foundation of open standards. This is central to WaveMaker’s AI roadmap. We decided to sit down with Prashanth Reddy, Senior Director of Product Management and Deepak Anupalli, Chief Technology Officer and Co-founder of WaveMaker, to shoot the breeze and find out their rationale for using AI this way.
WaveMaker: Prasanth, let’s start with how app development is generally evolving with AI and how AI is influencing developers. And, of course, where WaveMaker fits into all of this.
Prasanth Reddy: I would say we already know that “generative AI” is generating code, right? Everybody is saying, you know, when Microsoft Word arrived, all the typists employed in enterprises just got fired and their jobs went away. So, because AI is generating code, the programmer's job will also go away eventually, and with it, no code will also die. This is what people are thinking. I am generalising here for a reason. A lot of people who have seen technology demos think that way. But take custom or any application creation. It differs from just generating code for spooling out prime numbers, which is basically a single function. But we know that application creation goes through iterations, and different people collaborate, right? So there are people from the design team, there are people from the programming team. This is how enterprises work. The IT team is not in charge of ideating and creating designs. There will be a team that is ideating and creating designs, and then someone else makes a decision to take it to market.
So AI is all fine, but if you want to really adopt it within your enterprise, within your existing workflow of creating applications, you need a platform that allows existing teams to play well, and that is what AutoCode lets you do. Custom application creation needs iterations. You don't know everything up-front. You only have a sketchy idea of what you want to build, and what you want to take to customers ultimately may change as you go. You simply don’t have all the information. So first, you create something and implement it with, say, WaveMaker, and then iterate. That's how things get done.
Deepak Anupalli: So yeah, with this approach, Prasanth, are you indicating that one knows why one needs a low-code platform, that is to basically quickly put the pieces together, but now using AI? Is that what we are trying to do? Or are you saying that, okay, in the AI world you get some code but now how do you use this code, together with other bits of code, to create an application? That’s my first point. Two, now that you have assembled something, how do you iterate, how do you make enhancements? Whenever you make a change, a new piece of code is generated. How do you handle that?
WaveMaker: I think that is true, right? Maybe AI is not there yet in terms of providing you with a whole application, because an application is somehow not the sum of its parts, right? Yes, you can generate bits and pieces, but really putting it together and giving an experience? We probably are not there yet, are we?
PR: Yeah, for custom application development, the best expression of what you want to develop, the intent of what you want to develop, is a design. It can never be expressed as a spec. The whole ideation process, all the whiteboarding and everything, ends up as an artifact, which is a design.
WM: I agree but even before a picture, there are words describing the intent. There are discussions, there are sticky notes. Text prompts in a way.
PR: But at the end of it, there is an artifact that captures the intent, and that is not text, it is a visual. A picture speaks more than words, to use the cliche. So the information bandwidth that is communicated with a picture is much wider than anything you can do with text. So we are not simply what is possible with AI, instead what we are saying is that low code with generative AI gives you the ability to create custom applications faster using your existing teams, the way they are laid out, and your existing application architecture.
WM: So customers don't have to change anything to start using AI?
PR: Yeah, if you leave out a certain upper echelon of programmers – I challenge any typical enterprise IT or programming team working on custom app projects to have the agency and bandwidth to create and implement ideas. Most are “Okay, tell me what to do,” right? And that's where most of the custom application projects need attention.
WM: Isn't that what low-code had set out to do? A level playing ground among software developers, the democratisation of software development?
DA: No, wait. I think we are generalising too much here. Look at the different segments of applications. What does general low-code target? If you take typical citizen developers and business users, most want to automate routine tasks and execute faster. Naturally, many low-code platforms indeed make it easier to automate daily work. They focus less on design and more on automation because they were helping users who really didn’t want to go through piles of data and multiple screens to make decisions, but instead wanted to build an app that does it for them.
PR: So that is workflow automation that you're talking about. I think AI will consume workflow automation. In that space, one actually knows what one wants to build as the process is known ahead. AI can eat that lunch. But we are not caterers of that lunch.
DA: Yes, custom applications are where we are focused. It is important when we say building custom applications, we are assuming a well-defined design is available for it. But we are not talking about iterating the design. Yes, you are building a custom application but you are not creating a new design. Neither are you revising the existing design in your iterations, moving buttons around, changing an interaction design of a widget or whatever. So given a UX design, we have an AI way of arriving at a real working application. That’s what we are doing here.
PR: Both from a marketing point of view as well in terms of actual product offering, we do not have to necessarily stick to custom application development, though that’s the target use case. In general, WaveMaker is now AI-powered, inside out. It can not only convert working designs into working applications really quickly. You can further keep working on the code we generate – it’s open standards and all that. With Co-pilot and other AI, you either have to accept or rewrite the code that’s generated. You know you can't really work on improving that. Whereas with WaveMaker, we use AI to generate code but also give you a Studio where you can continue to tinker and continue to refine the code.
DA: I think this is where there will be value in WaveMaker, Prasanth. Typically iterative development is where you have people writing code and also modifying it, repeatedly. While AI gives you a higher ground to kickstart your app creation process, low-code makes it easier and faster to iterate. The big plus is the visual interaction model of the Studio which offers an intuitive way to fine-tune the experience – or refine the code.
PR: Yes. The code refinement process is frustrating with prompt interactions currently raging in AI. Nobody knows how to write the exact prompt, and the process of seeking it can be a long one.
DA: Currently, AI is a great start because it has the potential to take you from what you have imagined – your intent – to close to a real application very fast.
WM: Is it true that the flipside of the great start is that depending on how you write your prompts, it can expand the problem set rather than converge on the solution? While this may be good for research and ideation, it is not ideal in a production scenario.
DA: That's true.
PR: See that's why the expression of intent is best done with the visual design and not a prompt that nobody knows how to write. You may need to be a Linux command-line expert to grok that kind of thing.
WM: Even that may change. Thanks, Prasanth and Deepak, for another episode of WaveLength. Everything is pivoting faster than one can say pivot, so we must meet again and continue to discuss how AI-infused low-code changing developer experience in enterprise app development.