The rapid development of generative AI in traditional code development (third-generation language use) has had a lot of impact, with claims of massive productivity improvements. Given that developer productivity has historically been the domain of low-code tooling, this has led me to wonder whether the gap is shrinking and whether we are approaching a point where the benefits of low-code tools are being eroded for mainstream development.

To better understand this, let’s revisit how both technologies help.

AI-supported development

Delivered value in several ways:

• Code refactoring and optimization
• Code documentation generation
• Unit test generation
• Next generation of auto-complete

This can include creating code in a green field context. If you’ve been following reports on the value of services like Copilot, AWS Q Developer, and Code Assist, you’ll see that these tools are delivering a significant productivity boost. A recent ACM article pointed to benefits as high as a threefold boost for more routine activities, tapering off as tasks became more complex.

Low Code

Low-code tools have been around for a long time, while they have evolved and progressed, and have come in a number of forms, such as:

• UI applications that map databases to screens.
• Business process is defined with a visual tool support for BPM.
• Connecting different data sources by using visual notations to leverage representations of sources and sinks and link them together.

The central value proposition of low-code development is speed and agility. This performance comes with the constraint that your development has to fit into the framework, which may have constraints such as how it can scale, elasticity for rapid scaling, and performance optimization. ACM conducted some research into the productivity gains here.

Development acceleration narrowing

Low-code/no-code tools are often associated with the idea of citizen developers, where people with primarily a business background and a broad appreciation of IT are able to develop applications (personal experience points to more developers being able to focus less on code, and more on usability of apps). KPMG shares a view on this here.

Evolution of AI that could change low-code?

It would be easy to be a doom monger and say that this will be the end of highly paid software engineering jobs. But we have said this many times over in the last twenty or thirty years (e.g Future of Development).

Looking at the figures, the gains of Gen AI for code development aren’t going to invalidate Low/no code tooling. Where it really benefits is where a low-code tool is not going to offer a good fit to the needs being developed, such as complex graphical UI.

What if …

If Low-Code and Generative AI assistive technologies coalesce, then we’ll see a new generation of citizen developers who can accomplish a lot more. Typical business solutions will be built more rapidly. For example, I can simply describe the UI, and the AI generates a suitable layout that incorporates all the UX features, supporting the W3C guidelines. Furthermore, it may also be able to escape the constraints of low-code frameworks.

The work of developing very efficient, highly scalable Ui building blocks, with libraries to use them will still demand talented developers. Such work is likely to also involve AI model and agent development skills, so the AI can work out how to use such building blocks.

To build such capabilities, we’re going to need to help iron out issues of hallucination from the models. Some UX roles could well be impacted as well, as how we impose consistency in a user’s experience probably needs to be approached differently to defining templates.

Merging of assistive technologies

To truly leverage AI for low-code development, we will likely need to bring multiple concepts together, including describing UIs, linking application logic to leverage other services, and defining algorithms. Bringing these together will require work to harmonize how we communicate with the different AI elements so they can leverage a common context and interact with the user if using a single voice.

Conclusion

So the productivity gap between traditional development and low/no-code has shrunk a bit, I suspect we’ll see this grow quickly if generative AI can be harnessed and is applied, not just as a superficial enhancement, but from a ground-up revisit of how the load-code tooling works. Although the first wave, like everywhere else, will be superficial in the rush for everyone to say their service or tool is AI-enabled.