Using AI To Derive Application Intelligence And Drive Modernization

April 23, 2025 Marc Dallas

Everybody in the IBM i space knows that Big Blue is working on a coding assistant based on generative AI models to help developers write better and more consistent RPG applications with what everyone assumes will be a smaller and smaller programming talent pool.

But there is more to leveraging AI as a tool for application development and modernization than this. There are those who want to be able to use GenAI tools that can analyze and convert code from one language to another – much as the same large language models were first trained to translate between two languages and then across all languages.

There is an opportunity to use AI as a means of teaching both new programmers with very little experience as well as experienced programmers with lots of their own skills and context how to program – and program well – in RPG and COBOL, the languages of the IBM i platform. There are several large language models that can understand and generate pretty good RPG code, for instance, making them ideal as a foundation for tools that can teach RPG or COBOL up to a certain point – say over a four month crash course – and then follow that up with two months of human teaching at an expert level to learn the finer points of RPG and COBOL programming as well as whatever house style a given company has.

When ARCAD thinks about the use of AI for modernization, we think about the opportunity at two levels, at the code level and the application level. We know what you are thinking: Isn’t that the same thing? No, it is not. Code-level modernization is literally down at the line level, where you are either generating new code or explaining existing code to newbie programmers so they can understand what it means, how it works, and why the company chooses to code the way it does. With a particular dialect, so to speak.

At this level, ARCAD is very much interested in complementing IBM’s forthcoming code assistant, which is expected to be delivered later this year, with code explanation tools. In this case, one would take one of a number of AI models trained with code in various languages, use retrieval-augmented generation (RAG) techniques to pipe the company’s source code base into that LLM so the model could analyze and “codify” the house programming style. Such a tool might also suggest changes to that house style for performance or efficiency reasons, of course. A properly trained LLM could also give advice on syntax and how to create programs, but in the end, programmers need to have a sense of the architecture of the application suite – how it all fits together – and understand the relationships between programs and the different development strategies they embody. Some of these programs are two, three, or four decades old, and thanks to the brilliance of the AS/400 and IBM i system architecture and programming model (with its Technology Independent Machine Interface), many programs written so long ago just run on today’s hardware without any issues.

No other system in the world can say that. And no other system in the world ever will, either.

In any event, this hypothetical code explanation tool would work with IBM’s RPG code assistant tool in a virtuous cycle, improving the speed and quality of the new code created by actual programmers while maintaining house style, documenting the code effectively together with interactive explanation thanks to LLMs.

ARCAD is also very much interested in going one layer higher than the code by using AI tools to assist with application-level modernization, and by necessity it implies a hybrid approach. The reason this is true is that an application is never just its code. An application embodies an architecture and all of the services that plug into it. Metadata about these services is a key factor, and thus is where the ARCAD repository, ARCAD Discover, and ARCAD Transformer Microservices all come into play.

The metadata repository has what we call “application intelligence,” which is necessary to understand the architecture of current applications. You need to have a full grasp of the current architecture to figure out how to modernize it and, in most cases with very old code, make it more consistent and modular.

ARCAD Discover has an AI module that allows for the retrieval of architectural specifications and dependencies across application code and system services in a natural language – meaning the one that you and I speak conversationally. This is an important thing because business analysts, not programmers, tend to do the higher-level architectural work and bridge the gap between an application feature and a function in the business. These business analysts do not have the deeper technical skills that developers have to go poking around the system to find out the web of dependencies between code and services. But with ARCAD Discover and the ARCAD repository, they can come up with a modernization strategy and not rely on the memories of their developers, written down in lists and spreadsheets, to get a map of features and services as well as the source and compiled code that comprises applications.

To learn more about ARCAD’s use of AI in code and application modernization, please see:

Marc Dallas is vice president of research and development at ARCAD Software.

This content is sponsored by ARCAD Software.

This Issue Sponsored By

Table of Contents

Content archive

Recent Posts

Subscribe

Pages

Search