Low-code software development helped by AI

Artificial Intelligence (AI) has been getting a lot of attention lately. This is partly due to a number of technological innovations recently presented by the American tech company OpenAI. OpenAI drew attention to itself last year with the introduction of Dall-E, a generative AI model that can create unique images based on instructions in human language. A real hype emerged when OpenAI launched a new version of the intelligent chatbot ChatGPT on November 30 and made a free version available online. Within a few days of launch, more than 1 million users had already signed up for ChatGPT.

OpenAI's new AI applications have sparked many people's imagination about the possibilities of artificial intelligence. Following the enormous popularity of ChatGPT and Dall-E, Microsoft announced in January that it is investing billions in OpenAI. This has raised high expectations about the use of AI in existing software such as the online search engine Bing and MS Office.

What does this leap in AI technology mean for software development?

It's been common knowledge for a while now that developers are scarce and that this situation will continue for some time, while there's a growing demand for new software applications. Can AI offer a solution here? Can generative AI models like ChatGPT take the work of software developers off their hands?

The holy grail might be that anyone can be a programmer, and that you can ask an AI in human language to generate an application. We're still far from that. Yet the application of AI in software development is not a thing of the future, because it's already happening in modern low-code/no-code tools such as Mendix and Power Apps. In this blog you'll read which AI applications are already available in these tools and which innovations we can expect.

AI-driven tools in low-code tools

We already saw the first applications of AI and ML (machine learning) in low-code tools back in 2018, when Mendix introduced the smart assistant Mendix Assist. Mendix Assist helps developers model application logic in so-called microflows by continuously suggesting improvements and adjustments as a next step. Mendix Assist uses historical and anonymized data based on the many application models that customers have developed on Mendix's cloud platform. The Mendix Assist AI model recognizes patterns while you're modeling and gives you contextual recommendations for a next step, after which you can also have a suggested recommendation carried out if you accept it. This way, you not only build better applications, but developing them also goes faster than if you had to do it yourself.

In the years since its introduction, Mendix Assist has been expanded with new features, including for creating pages (screens) and automatically generating validation flows. Besides speeding up the development process, Mendix Assist nowadays also helps optimize the performance of applications. The Performance Bot within Mendix Assist checks an application on all kinds of aspects during development and detects potential design flaws in the data model, the microflows, and the pages of an application. Detected issues in the application's design are clearly explained to the developer and can then be automatically resolved by Mendix Assist.

Microsoft launched Express Design for Power Apps in May last year. With Express Design, an existing wireframe (design) from Figma, an image of a paper form, and even a screenshot of a whiteboard can be automatically converted into an app. Power Apps' cognitive AI model recognizes the structures in an image and can translate this into a working user interface for a new application. The model is even able to create a data model in Dataverse based on the image.

During the latest Ignite event in October last year, Microsoft presented an AI innovation for Power Automate that gives us a glimpse of what to expect in the near future. Here, a preview feature called Create a flow from a description was presented, which lets you describe what you want to automate via a command bar. The entered text is then translated into code by Power Automate. Based on the code, a flow is automatically modeled for you with the necessary activities and parameters to carry out the requested automation. Power Automate uses OpenAI Codex for this, which, like ChatGPT, is a generative AI model for generating code.

Embedding AI functionality in low-code apps

Another trend we see in low-code software development is the integration of AI functionality into applications. Think, for example, of connecting an application to standard cloud AI services for automatic image recognition, or converting audio to text. These services can be connected fairly easily with a low-code tool. In addition, many companies are working on developing custom AI models for specific applications, such as process optimization, fraud detection, and predictive maintenance. Embedding this kind of AI model in low-code applications was difficult until recently, partly because complex APIs and REST services are needed to call the models from a low-code tool. Recent innovations in Mendix and Power Apps can solve these problems.

Microsoft has offered the ability to create AI models and use them within Power Apps through AI Builder for some time now. In AI Builder, you can choose to use a pre-built model, or build your own model in a low-code way (without programming). AI Builder offers templates for various model types, including category classification, document processing, and predictions. Microsoft has recently expanded the capabilities of AI Builder, so you can now import custom ML models developed with the Azure Machine Learning Platform into AI Builder. Once an AI model has been created in AI Builder, the model can be called within Power Apps using Power Fx. This is a low-code formula language from Microsoft that works in roughly the same way as Excel.

Mendix follows more or less the same example as Microsoft with the new Machine Learning Kit, which was announced at the end of last year and will officially be released very soon. This should also make it possible to add a custom ML model to an app, making it much easier to call the model within Mendix.

What can we expect?

So far, it's mainly smart AI-driven features in low-code tools that make developing applications easier and that produce better software with these tools. AI acts as a kind of co-pilot, taking tasks off a developer's hands and protecting the developer from making design mistakes. In the near future, I expect more of these kinds of applications, where the developer is helped not only with building applications but also with other aspects of software development, such as automated testing of applications.

With the recent technological developments in the field of generative AI, I expect we'll see new AI applications in low-code tools this year. With the investment in OpenAI that Microsoft has announced, it could well be that a smart AI chatbot gets integrated into Power Apps and Power Automate. The first signs of this can already be seen in preview features such as Power Apps Ideas and Create a flow from a description in Power Automate. This makes it possible to ask, in human language, for a specific part of an application to be modeled. Now that's what I call next-level low-code app development!

Will software developers become obsolete because of AI?

Will flesh-and-blood software developers eventually become obsolete because of AI? I don't believe that. If code is automatically generated by AI, you as a developer still need to be able to review and adjust the code where necessary. We can't rely on the output of generative AI models always being correct, or that it precisely matches the intentions of the client. Moreover, software development consists of much more than code. Having conversations with users and translating user needs into functionality that can actually be built remains work for people. The expert role of an experienced developer or a tool specialist therefore remains necessary in software development.

Want to know more about low-code?

Want to know more about the possibilities of low-code tools, or do you have a concrete challenge that a low-code app could solve? We're happy to help with advice and to think along about the possible applications of low-code for your specific situation. Get in touch with one of our account managers for an introductory conversation.

Get in touchSchedule an appointment

Written by Steven Samuels Brusse,
Senior Consultant