Will AI Code Generation Replace Low Code?

AI code generation is still wowing the software development world. 

Google recently unveiled its Gemini offering, estimating that the AI model could outperform around 85% of participants in competition coding. And I estimate that it could outperform slightly more participants than that if I happened to sign up for the competition.

There’s a lot to be excited about when it comes to AI in app development. 

As the tech proliferates and improves, we could see more and more non-developers using intuitive tools to create complex software. Meanwhile, professional devs can cut development time by automating tedious structural code and focusing on more creative work.

But wait, I’ve heard these benefits espoused before. And I don’t just mean the many, many, many times I’ve written about AI coding assistants recently. Solutions that seek to democratize software development by automating basic code chunks have been around for some time now. 

In fact, low-code development platforms have existed in some form since the mid-2000s. Yet the collaborative potential of that software has arguably never been fully realized on a wide scale. 

What makes the promise of AI code generation different, if at all, and will it outright replace low code?

AI code generation and low-code: different methods, same result?

It’s easy to look at AI code generation and low code and say that they’re essentially achieving the same thing. Users employ a natural input method to build features and applications, ideally with minimal manual coding. However, there are key differences between the two technologies. 

AI code generators receive natural language prompts and output code based on the prompt. 

For example, one might type, “Write me code that will scan PDF documents for instances of the word “duck” and output the final count.” 

The AI code generator would then interpret the desired functionality and write code on the fly that attempts to achieve that functionality. Yes, the AI model relies on its training data to perform its output. But the outputs it’s achieving are unique according to the prompt it receives.

Given that this tech is relatively new, the implications are enticing. Theoretically, the limitations of the generated code lie only in the creativity of the user and the capabilities of the AI model. And AI models are getting better at interpreting prompts and generating complex code.

Meanwhile, low-code methods typically automate coding tasks via preset, visual abstractions of code. 

For example, one might drag a prebuilt “search bar” element into their application and define some parameters from there via a drop-down menu of settings for that element. The addition of the search bar itself, as well as the fine-tuned settings, all change the underlying application code without the user having to understand that code. 

The difference compared to AI code generation is that traditional low-code tools typically aren’t generating code on the fly. Instead, they rely on a series of rules and configurable chunks of code that are then represented to the user as visual application elements.

For now, that means that these two technologies still have distinct use cases. AI assistants generate raw code, but they typically don’t abstract that code into visual elements. Many of them function within the development environment and act as a “helper” throughout the coding process. That means they work best for users who understand how to implement and adjust the generated code. 

For experienced developers, these tools are a godsend for productivity. The potential for customization is higher than that of low-code, but the usability is a bit lower. Contrast that with low-code platforms’ configurable, visual application builders that “just work.” Users with less experience or who don’t require highly custom functionality to achieve their end goal will find low-code tools more appropriate for their needs. 

Low-code is still the dominant space 

The G2 review data for these categories tell an interesting story. 

Low-code and no-code have been popular categories for a long time, and that’s reflected by their relatively high average monthly review counts. AI code generation is new on the scene, but its monthly review count is holding steady. 

Comparing average monthly review counts from Q3 2023 to Q4 2023, the gap between low/no-code and AI code generation is actually closing. 

That indicates that while the tech still has some ground to cover in terms of popularity, AI code generation is proving itself as a useful and relevant way to build software quickly. 

The takeaway

It’s hard to say whether we’re on the cusp of a complete paradigm shift regarding non-manual coding or whether that’s still a ways off. AI code generation has definitely distinguished itself from low code. 

But while the hype is real, it’s still decidedly the less popular technology. 

However, I do predict that code generation will eventually become as prolific among buyers as low-code, if not even more so. The potential for code customization with AI tools is too great to ignore. 

What’s missing is buyer awareness and usability, especially for non-developers. Soon, we could see more low-code platforms integrating AI functionality to give users the best of both worlds. Salesforce and Outsystems, the respective #1 and #2 vendors in G2’s Low-Code Development category at the time of writing, already do just that. I predict that AI code generation will change the concept of low-code itself or outright replace it. The question is when, not if. 

Curious about AI code generators? Uncover the legal essentials before you delve in and make informed choices for your AI coding endeavors.

Edited by Sinchana Mistry