The future of GenAI will rocket fuel modernisation of core legacy systems

According to our research, forward-thinking firms are driving the adoption of emerging tech. 

Specifically, 87 per cent of business and technology professionals who have some knowledge of emerging technologies are experimenting with, piloting or implementing artificial intelligence. 

The third most common use case for generative AI is software development.

Many are enthusiastic about the chances of writing code in natural language or using an autocompletion editor, or pair programmer, that helps directly write code quickly.

Financial services organisations like ANZ and Westpac have run internal experiments to prove that software coding can increase productivity between 42 per cent and 45 per cent. 

However, there is much more that GenAI, as well as AI and machine learning, can collectively do in the world of software development and maintenance.

Forrester has named the area of AI-enabled assistants for software development Turing Bots, in honour of the British AI scientist Alan Turing. We consider Turing Bots to be a new exponential technology that will have short and long-term positive impacts in the financial services industry.

Turing Bots speed up the software development lifecycle

Turing Bots can support all phases of the software development lifecycle (SDLC), from the initial planning, analysis, and requirements phases, to design, coding, testing and delivery. There is a TuringBot type for every phase of the lifecycle.

Indeed, a rich eco-system of tools built by vendors is developing fast in the market. Well-known software and platform vendors like GitHub, Atlassian, Google, and AWS, together with start-ups and smaller players like Applitools, Coedium, Tabnine, as well as the well-known OpenAI, are all racing to build and infuse Turing Bots in current and future development tools. 

More advanced Turing Bot approaches are using open-source large language models (LLMs) specialised in software as data, including Codex, Code Llama, and StarCoder. The technology is getting there, but are we ready for it? No.

As technology evolves faster and faster, we humans and enterprises struggle to keep up and change the way we work. Considering enterprise resistance to change, we believe that in less than five years, this technology will totally disrupt the way software, applications and products are built and maintained. 

What about legacy?

When discussing legacy technology, let’s use ATM machines as an example. Interestingly, 95 per cent of ATM card swipes to withdraw cash run on old Cobol programmes. Cobol stands for common business oriented language, and is a computer programming language that has primarily been used in business, finance, and administrative systems since 2002.

It is estimated that more than 220bn lines of Cobol code are still used in production around the world. 

This is the beginning of a new world, where the cost of migrating legacy systems to modern technology stacks is no longer prohibitive.

Legacy applications process, run and manage customers’ policies for insurance products, and hold much of the data that is crucial in keeping business operations up and running. That data is also crucial for modern applications, but access to it through legacy applications is not easy. 

Legacy does not just mean Cobol. Even applications built with newer programming languages like Java – including older versions such as Java6 – are now legacy. Dealing with legacy is a challenge that has been slowing modernisation for years. The attitude towards legacy has always been, 'we have no clue how, but it works, so we better not touch it'.

TuringBot use cases can also deal with legacy

That is all about to change because GenAI offers great opportunities to help the modernisation of any type of legacy, as long as LLMs have been trained or refined on the specific legacy code.

The good news is any coding Turing Bot that has learned how to write and understand code in various programming languages can be refined and customised to understand code of a new language.

Coder Turing Bots – or AI-enabled assistants that generate code from English language input – all work with the top 10 (and more) trending programming languages, including Java, C, C++, Python etc. They can help do many things that we could not do in the past with legacy, at least not without exceptional effort and time. 

Here are some of the most relevant use cases: 

• Automated documentation generation for a legacy programme file. Documenting legacy is the first step to unlocking opportunities in modernising old legacy systems. While this does not, and cannot, all happen at the push of a button, there are real experiments going on at scale in large global organisations.
• Automated translation from one programming language to another. Most coder Turing Bots can translate old Java code files into more recent Java code, but they can also translate Cobol into Java. For example, IBM is enabling WatsonX.ai with a 20bn parameter LLM for code to deal with Cobol.
• Automated test case generation and optimisation from requirements. Test cases are fundamental in testing that the original requirements implemented by the legacy system are also met by the system that has been migrated to the modern language. Tester Turing Bots assist with this use case.

An example of automated migration from Cobol with Turing Bots 

The above use cases are already helping a lot in a large system integrator’s experiments in migrating from Cobol legacy to modern architectures. In a nutshell, code LLM-enabled Turing Bots are being used to generate documentation from Cobol (and JCL) programmes.

The documents are reworked into requirements and an overall understanding of the application architecture, which is an essential starting point for dealing with any legacy code.   

Furthermore, GenAI infused in tester Turing Bots generates test cases from the documented requirements. In parallel, WatsonX.ai translates Cobol and JCL into modern Java code.

Next, the newly-generated Java code gets tested manually and automatically with the test cases. This process guarantees that the new code meets the functional requirements extracted from the Cobol.

This semi-automated process is a starting point; development teams can enrich it further with additional steps, like adding new requirements or capabilities during the document generation process. In other words, further re-engineering of the old application or business process is possible throughout the overall process.  

This is the beginning of a new world, where the cost of migrating legacy systems to modern technology stacks is no longer prohibitive.  

In the near future, organisations will go from not touching legacy to migrating or regenerating it into new and more modern applications.

So, if you are one of the many companies stuck with legacy systems that you can neither afford to migrate nor replace, do not despair. GenAI has the potential to radically reduce the cost of migration.

Diego Lo Giudice is vice-president, principal analyst at Forrester