"If generative AI has advanced this far, then low-code development may no longer be necessary."
If you're working in an IT department and promoting digital transformation, you've probably considered this at some point. Generative AI is not only used for development in natural language, but it's also becoming established in the form of "AI agents" that actually perform tasks.
In short, low-code is necessary. As will be discussed later, it is especially indispensable for the digital transformation (DX) of enterprise companies. Low-code is a method of developing applications primarily using a GUI, minimizing manual coding. However, nowadays, there are even platforms that support everything from planning and coding to operation and maintenance after application implementation on a single foundation. These "low-code platform technologies" are characterized by their ability to provide both speed and governance for enterprise companies.
The question that should be asked is not whether to choose generative AI or low-code, but how to use them effectively. Generative AI is responsible for executing tasks, while low-code is responsible for creating business applications that incorporate that generative AI, maintaining control, and connecting them to existing systems. In an enterprise setting, the two are not competitors, but rather have a division of labor.
Macnica, a trading company specializing in cutting-edge technologies, implemented this division of labor by reforming its own quotation process. This article will explain how Macnica streamlined its operations and why it chose low-code development in the age of generative AI.
This article is a reconstruction based on a presentation given by Mr. Tone of Macnica 's Innovation Strategy Division.
AI agents have transformed the quotation process to this extent.
In one of Macnica 's business units, the quotation process was a major headache. The Excel quotations received from customers varied widely in terms of required fields and formats. Salespeople and operators who received them had to manually re-enter the information into the system. Naturally, omissions occurred. In some cases, quotation data was not even kept within the company.
The manager from the business division brought this issue to the attention of the “CoE (Center of Excellence),” an internal organization that oversees all company-wide DX initiatives at Macnica. The CoE at Macnica has received over 70 ideas in a single year, and has already released nine operational improvement systems and two new services. This issue with the quoting process was one of them.
Further details on creating this Center of Excellence (CoE) organization are explained in the following article.
At the time, this manager lacked cutting-edge knowledge of generative AI. Therefore, the project began with the CoE members explaining "what AI can do now." Through continued communication between the manager and the CoE, it was decided to proceed in the direction of "entrusting tasks previously judged by humans to AI agents."
In the implementation, three agents will share the work.
The first robot's job is to extract information from Excel spreadsheets with varying formats. It takes over tasks that previously relied on human judgment.
The second unit is responsible for automatically transferring data to the company's internal system. Different rules for each customer are handled by adjusting the prompts.
The third person is in charge of checking the answers. They determine which parts of the agent's answers are correct and which parts are wrong and why.
It's difficult to completely hand over tasks to an AI agent at this stage. Therefore, assuming that 100% accuracy won't be achieved, we designed it as a Human-in-the-Loop system (integrating human verification into AI processing), relying on the results of a third AI's verification and having a human make a final confirmation. This extra step prevents errors in registering unit prices and product names.
This system significantly reduced my work time. Furthermore, being freed from the stress of endlessly comparing Excel spreadsheets was a major benefit.
Originally, at the initial drafting stage, the idea of "using agents" was not something the business divisions had considered. This To-Be (ideal state of operations) emerged through repeated discussions between the business divisions, who were familiar with the operations, and the Center of Excellence (CoE), who were familiar with the technology.
How I differentiated between using generative AI and low-code: The point of divergence from scratch code generation AI.
We can turn something that is difficult to share into a tangible form within 5 days of the initial proposal.
So, how did they bring this "To-Be" vision to life? The answer is that they quickly created something that gave a clear image, and then proceeded by aligning that image with the business divisions and the Center of Excellence (CoE).
In this case, Macnica, with just one developer and within five days of the project proposal, created a mock-up of an AI agent for estimation tasks. If it wasn't quickly realized, the priority of the project would steadily decrease for the busy business departments, and expectations for digital transformation wouldn't rise. That's why speed is so important.
There's a reason why we insisted on such meticulous adjustments.
For business units, AI agents are unfamiliar, and they lack a clear vision of how to use them in their work. Furthermore, the business units' desires are often highly abstract, making it difficult to translate them into requirements without discrepancies. Moreover, even if an AI is created, it may not be used in the field. To prevent this, it was crucial to develop using an agile methodology, aligning ideas while working with the actual product.
At Macnica 's Center of Excellence (CoE), multiple tools such as Power Platform and UiPath are used depending on the nature of the project. For this particular project, Mendix, a low-code development platform strong in "sharing ideas" and "speed in bringing projects to fruition," was chosen. The three AI agents seen earlier are integrated into and running on a business application created with Mendix.
From the available options, Macnica made its selection based on two criteria.
Looking at the market, low-code continues to grow, and the next point of discussion is shifting to its combination with AI.
Mendix is part of this trend. Mendix is a low-code development platform that enables IT and business departments to collaborate on a common foundation to build enterprise-level core systems. In recent years, it has evolved into an AI integration platform combining "low-code × agentic AI" by integrating with data analysis and AI platforms.
So, what did Macnica choose at the time? Besides Mendix, other options included development from scratch, Power Platform, and the much-discussed code generation AI (such as Claude Code). The code generation AI was intended to handle not business processes, but system development itself.
Macnica based its decision on two main criteria. First, as mentioned earlier, "Can we quickly bring the vision to life while sharing it?" Second, they emphasized "Can we meet enterprise scalability and control requirements?" Scalability means being able to connect and expand with existing internal systems, such as core systems. Control means being able to manage development and operation under company-wide security standards and operational rules. From this perspective, Macnica 's assessment was that none of the options mentioned above would be feasible at this point.
With scratch development, immediate screen-based review is not possible (a barrier to image sharing and speed). Power Platform has inherent limitations when connecting to systems other than Microsoft 365, as in this case (a barrier to scalability). Code generation AI can do many things now. However, code generation AI is only responsible for the coding part of application development. Enterprise applications require mechanisms outside of coding, such as separation of development, testing, and production environments, operation of the execution environment, and user access control. This is outside the scope of code generation AI, and all of this must be built and operated in-house (a barrier to control).
We developed an in-house application using Mendix that also integrates with our core systems.
Regarding the second axis, "scalability," Macnica has a proven track record of developing applications that integrate with core systems using Mendix.
Macnica 's IT department replaced the expensive subscription-based SaaS they were using to manage their IT budget with an in-house application developed using Mendix. This application required integration with SAP, but it was released as a standalone product and is still in operation today.
Low-code development isn't limited to simple apps; it can handle even full-fledged applications that integrate with core systems. This is the strength of Mendix 's "scalability."
| choice | Axis 1: Image sharing and speed | Axis 2: Scalability and Control | Macnica 's assessment |
|---|---|---|---|
| Development from scratch | × (Unable to immediately write a review while looking at the screen) | − | Image sharing and the speed barrier |
| Power Platform | ○ | × (Restrictions on connections with non-Microsoft 365 systems) | Scalability barrier |
| Code Generation AI (Claude Code etc.) |
− | × (Ensuring company-wide governance and security is a challenge, including in terms of organizational structure.) | Control wall |
| Mendix | ○ | ○ | Recruitment |
However, tool selection is only "one-fifth" of the process.
While the two previous examples were successful, Macnica also went through periods where things didn't go well.
Previously, each department used disparate tools and pursued development using its own unique processes. This was what's known as a "rogue project" situation. In this state, rapid transformation that leveraged the strengths of the entire company was not possible.
To break out of this situation, Macnica adopted the "5P" concept, originating from Siemens, the parent company of Mendix, as the key to its progress. The 5Ps are a system that compiles the vast amount of practical knowledge of digital transformation in Europe and the United States into best practices, organized into five elements: Platform, People, Portfolio, Process, and Promotion.
"Tools are only a small part of DX transformation." In other words, the Platform (tools) is just one of the five Ps, and the transformation only progresses when People, Portfolio, Process, and Promotion are all in place.
Macnica has put this idea into practice, and the Center of Excellence (CoE) organization was born from this concept.
*The "Digital Execution Factory" shown in the diagram above is a DX support service for the manufacturing industry that includes support for building a CoE organization, in order to generate continuous DX.
The Center of Excellence (CoE) serves as a point of contact for receiving ideas for company-wide digital transformation (DX) and new businesses. It plays a role in producing outcomes such as business process improvements like the estimation work mentioned earlier, and new services. To this end, it has established a development environment and planning and development processes. It selects recommended tools and distributes templates and security modules that conform to the company's design standards to all projects. The "control" aspect, which was included in the second axis mentioned earlier, is a combination of the security and environmental management functions provided by the platform and this system for operating them company-wide. Mendix also possesses the know-how to build this system, which has been extremely powerful.
In addition, we will provide support for developing DX talent and promote projects. Furthermore, we will gather the know-how gained from all projects in one place and utilize it for future projects.
In other words, it's about building a system that utilizes all 5Ps, not just the tools. This is what allowed us to create the aforementioned example.
Furthermore, there are certain types of "P" (Property) that tend to be lacking depending on the type of DX promoter, and these often become obstacles to DX implementation. For more details, please see this article.
Why low-code in the age of generative AI?
Macnica itself used to outsource system development to external partner companies. However, with the advent of generative AI and other factors, this approach has become obsolete in this era of rapid business change.
To keep up with changing times, the IT department needs to quickly visualize the highly abstract tasks held in the minds of business units and then concretize the processes through a process of alignment. The approach we demonstrated in the example of estimation work is necessary.
This is where low-code plays a crucial role. It allows business and IT departments to align their understanding and quickly bring ideas to life, while also meeting enterprise scalability and control requirements. Based on these two criteria, Macnica chose Mendix.
Code generation AI will continue to evolve, and its capabilities will increase. Low-code development, on the other hand, will also evolve by incorporating generation AI.
Low code is still necessary in the age of generative AI. In fact, precisely because of the age of generative AI, it's crucial to quickly create AI that performs tasks, deliver it to the field while maintaining control, and integrate it seamlessly with existing systems. That is Macnica 's answer to the initial question, "Why low code?", which we've learned through practical experience.
Furthermore, the fact that so many successful cases have emerged within Macnica is due not only to Mendix, but also to the 5P know-how and the organizational structure that enables its implementation.
When trying to continuously promote DX on a company-wide scale, even if an extremely advanced AI is created, it alone will hit a wall. No matter how excellent the AI is, in terms of the 5Ps, it is only one of the Platform (tools). Introducing tools is just the beginning; what is truly important is the remaining four-fifths.
And Mendix has the know-how of these 5Ps.
Through its own practices and experience supporting numerous clients, Macnica has realized that following best practices can lead to highly reproducible DX implementation. Considering not only improved development efficiency but also the long-term sustainability of DX, choosing Mendix was an excellent decision for Macnica.