EvolveWare Blog

AI-powered Legacy Application Modernization: Separating Fact from Fiction


With the announcement and release of ChatGPT, artificial intelligence (AI) and more specifically Generative AI (GenAI), was thrust into the limelight. Expectations for what the technology could help us achieve soared to new heights and left many wondering if the technology could become the “silver bullet” that many have hoped for, particularly for complex and time-consuming tasks such as legacy application modernization. In fact in a study run by IBM, 79% of business leaders said they believed that generative AI tools for application modernization would significantly improve business agility. Of course with every “yay-sayer”, there were equally those who urged caution and advocated for a measured approach to integrating this technology into application modernization workstreams and processes. So now that we are about a year and half in, what does the outlook seem to be? Can AI or GenAI truly be a gamechanger for application modernization initiatives? We believe that answer starts with understanding the history around AI and app modernization.

The History of AI and Application Modernization 

GenAI is not the first AI technology to be leveraged for app modernization initiatives. Machine learning models and supervised AI methodologies have been used both for understanding the components of source applications as well as transforming code from one language to another. The benefits of using these technologies to assist with tasks involving pouring over thousands if not millions of lines of code are well-documented. For example, in tests conducted with partners over other manual methods and tools, EvolveWare’s Intellisys platform has reduced time and effort spent on generating documentation artifacts for an application by 90% or more using its unique machine-learning based technology. 

However, challenges have also occurred. For example, one popular method of speeding up application modernization initiatives was to simply take legacy code and convert it line by line to code in a modern language using a transformer or neural network architecture. COBOL to Java conversion was the most popular conversion of this kind, but the two programming languages are inherently different. And most transformers didn’t take this into account. So the code that an organization was left with in most cases was aptly termed “JOBOL” – Java code that still looked a lot like the original COBOL code and contained all the original flaws and vulnerabilities. Because of this, organizations who used this method were not able to realize the gains and efficiencies from modernizing that they anticipated. Now, some companies, such as ours, do offer the ability to transform code more contextually and to optimize the code prior to migrating it, alleviating some of the concerns of line-by-line migration. The generated code is about 75-90% of the way there and does require some level of manual work post transformation to  get to a completed target application.

In a nutshell, AI technologies can without a doubt help speed up and de-risk these efforts. But the particular methods used to train and implement these models as well as the maturity of these technologies do factor into the success rates. That is what we need to remember for this wave of the hype (and potential) around AI. 

The Potential of AI Today (And a Few Missteps to Watch For)

The story around AI today primarily focuses on GenAI as the “new kid on the block”. The potential to use this technology for legacy application modernization initiatives lies primarily in its ability to translate – both in terms of better explaining current applications as well as the ability to convert the programming language into modern code. In fact as part of its 2024 Predictions for IT Organizations and Users, Gartner stated that by 2027, GenAI tools “will be used to explain legacy business applications and create appropriate replacements, reducing modernization costs by 70%.” 

 However, organizations should be careful about using this prediction in a vacuum. As evidenced from the history of AI and application modernization, simply using GenAI techniques to convert code will not result in the efficient and scalable modern applications that organizations need. When it comes to AI, the data used to train models is crucial to achieving success. Ensuring that the data used is robust and representative of the relevant languages as well as providing the ability to assess and optimize the code prior to conversion is important to achieve the organizational goals behind migration. Similarly, the data used must also be taken into consideration for models that help with use cases related to understanding code. Most legacy code resides within the long-standing organizations that have used these languages instead of being widely and openly available. So for GenAI models to be successful for these app modernization initiatives, they must be trained on these organizations’ data. And putting the proper security and use protocols in place and experimenting with these models will take time.

But what about other AI technologies? With all the chatter around GenAI, many longer-standing technologies such as optimization, rules and heuristics, and non-generative machine learning have fallen to the wayside. But there is still plenty of potential, if not more immediate potential, that remains for these technologies to speed up processes involved in legacy modernization. What’s most important is to find the right use cases based on the time-consuming parts of these initiatives. For example at EvolveWare, we are currently looking to incorporate additional seasoned AI methods to speed up processes involved in extracting the business rules from an application, such as suggestions for rule categorization and rule summarization. And we do see the potential of using large language models (LLMs) associated with GenAI for outputting more natural language business rule summaries – with the right data, modeling and training of course.

AI Can Help…But Its Not a Silver Bullet

As we’ve seen, AI technologies both today and in the future can help speed up processes involved in modernizing legacy applications. There is quite a bit of potential as we continue down the path. However, these technologies are not a “one-click” or “silver bullet” solution. Instead they should be thought of as intelligent assistants – giving us as humans the ability to process and complete tasks with more lines of code than we could handle on our own but not completely replacing the need for our involvement in these projects. And as intelligent assistants, these technologies require the right kind of inputs, training, and implementation in order to help us ultimately achieve success with our organizational goals.

To learn more about how EvolveWare is using AI to speed up and de-risk the application modernization process, contact us here.


You May Also Like…


Submit a Comment

Your email address will not be published. Required fields are marked *