Breaking the code: The power of GenAI and model-driven software

Recently, I wrote about the risks created by low code platforms that are using generative AI as code or script generator. Inspired by this post from Forrester’s John Bratincevic and Diego Lo Giudice, I wanted share a little more about how I think low-code platforms should be using GenAI.

Low-code is itself a form of “model-driven development” – a concept that has been around for a while. In fact, Pega was founded on the idea that authoring workflows and business rules as models – not code – was a better way to build business process software. What does that mean? A model is simply a visual representation of a real-world thing or concept. For example, in computer aided design (or CAD), the object to be built – maybe a whole car, or maybe pieces of the car – starts as a model in a 3D, visual tool. 3D printing takes this concept even farther – you can create a visual model on a computer and with just a few clicks, turn it into a physical object you can hold in your hand.

Model-driven software is like CAD-powered 3D printing for your business. You can draw a business process as series of stages and steps, and with a few clicks, turn that into a running piece of software that manages and automates that process. You drag and drop fields on a user screen and have that instantly become a usable web or mobile form. You can define business logic in the form of tables or decision trees and be able to run those rules with real data immediately. Put enough of these pieces together, and you can build a full functional decisioning and workflow application without ever writing a line of code.

If the structure of the model is robust enough, you don’t need to have generative AI write code. You can have generative AI build the models directly. You can ask Gen AI to lay out the steps of workflow simply from a name. You can have it suggest data models for the workflow, and then map map data from a 3rd party REST service into that data model.?

There are three major benefits to this approach:

• First, you get visibility and maintainability. In a low-code, model-driven approach your business logic isn’t buried in code – it’s in visual models that can be easily read, interpreted, and changed not just by professional developers, but by subject matter experts – citizen developers – who bring a deep understand of the business objectives. It also reduces the risk associated with using generative AI inputs: since users can clearly see exactly what the LLM has suggested, they can easily review or override. And as business needs change – which they inevitably will – users can quickly change their business processes.
• Secondly, you get reusability and standardization. Model-driven approaches enforce componentization. Each model is “fit for purpose” and has implied inputs and outputs – essentially an API signature. By defining commonly used software components as models, such as decision trees, decision tables, and service level agreements, developers can create a library of reusable assets. These assets can be easily shared and reused across multiple projects, ensuring consistency, reducing development time, and minimizing the risk of errors. Some low-code platforms have developed patented architectures which maximize reuse while allowing an enterprise to specialize business logic to the unique needs of a product, region, or customer segment.
• Finally, you get speed and agility. Leveraging the output of a large language model can expedite the development process by automatically generating the initial structure and model of an application.??and code snippets based on the models provided. Because the low-code model defines exactly what is needed for a complete application (workflow steps, screen definitions, data models, etc.), the low-code platform can prompt the LLM for exactly the information needed to build out the structure of a complete app. Rather than trying to make a bunch of code snippets work together, you get a collection of models that designed to work together from the start. You can actually see how fast you can get started with this approach.

There are other benefits to a model-driven approach, including increased security, consistent application performance, and better collaboration between business and IT. As you look for ways to harness the power of generative AI to create immediate business value, using LLMs to accelerate your low-code development is a low-risk, high-value place to get started.

Just don’t let them generate a bunch of code…