For more than three decades, a leading automotive manufacturer relied on a mainframe system as the backbone of its sales process. But this massive, complex architecture, composed of 15 million lines of COBOL code and an integrated database management system (IDMS), was becoming increasingly fragile. With many of the engineers who fully understood it nearing retirement, the pressure to modernize was on.
To ensure it could continue to deliver value into the future, the company launched a priority program to rebuild this core system by migrating it to a microservices architecture. But it was a daunting task. They needed to rewrite the entire codebase and move off the mainframe by the end of 2025 to avoid licensing issues.
With the program running behind schedule, the automotive giant turned to Thoughtworks for help.
The reverse engineering bottleneck: Unpacking 15 million lines of code
Before Thoughtworks’ involvement, the company had identified the reverse engineering process as the main cause of delay. The average lead time to reverse engineer 10,000 lines of code was six weeks, using two full-time employees (FTEs) and a subject matter expert (SME) for review.
With COBOL engineers becoming increasingly scarce, and fewer younger developers learning the language, expertise was hard to find. While the research and development team hired third-party developers to reverse engineer the code, only a handful had specific contextual knowledge of the manufacturer or the unit’s domain. Meanwhile, business analysts were struggling to get useful information from technical documentation and were forced to rely on SMEs, who were already spread too thinly across multiple teams.
In an effort to accelerate the program, the company explored using generative AI (genAI) to reverse engineer the legacy codebase. But the genAI outputs didn’t meet the business analysts’ requirements, and technical SMEs found they were only 60% accurate.
Solving the SME shortage with an AI breakthrough
The Thoughtworks team began by running a value stream mapping session to analyze the end-to-end reverse engineering process. This provided deep insights into overall efficiency, resource allocation and major bottlenecks — the most significant of which was the limited availability of SMEs.
We recommended that SMEs should spend less time reviewing inaccurate reports and more on providing suggestions for forward engineering design and the “to-be” state. This would help business analysts and SMEs to work faster and support multiple workstreams, helping get the program back on track.
To test this, we ran a proof of concept (PoC) using Thoughtworks’ legacy modernization accelerator tools, which integrated with our client's AI platform to automate document generation and provide a chatbot to answer questions. Throughout, we worked closely with the business analysts to ensure they got the critical technical information they needed.
Today, these legacy modernization accelerators form part of AI/works™, our agentic development platform, which helps organizations address the complex challenges of legacy modernization by providing a deeper understanding of existing systems and reducing SME dependency.
A 66% reduction in lead time
During the PoC, we also conducted an experiment to enhance the tool’s code comprehension pipeline using abstract syntax trees (ASTs) and genAI, with an aim of generating documentation faster and with more clarity.
The results were immediate — what would have taken two FTEs more than four weeks could now be produced in just a few hours. These results led to a second phase to productionize our legacy modernization tool for the client.
We also tested the chatbot’s ability to manage the requirements clarification questions that business analysts typically asked SMEs. Using ASTs, the tool was able to extract logical units of the code (such as packages, classes and functions) and visualize how they related to each other using a knowledge graph.
With this improved database, the retrieval-augmented generation (RAG) system found more relevant information about the code, leading to clearer and more helpful answers and considerable impact. Reverse engineering time was reduced by two-thirds — from six weeks to around two weeks for each 10,000-line module.
This represented a potential saving of 60,000 person-days for reverse engineering all 15 million lines of code.
Our legacy modernization tool slashed manual effort — freeing teams to focus on higher-value work.
Reverse engineering cut from six weeks to two per module.
Massive efficiency gains across 15 million lines of COBOL — driving real business impact.
Scaling modernization with fewer blockers and more confidence
By accelerating reverse engineering and freeing SMEs to support parallel workstreams, the PoC demonstrated how the company could get this major mainframe modernization program back on track. Its success gave them the confidence to use AI to accelerate the program, knowing that the outputs significantly helped business analysts rather than hindering them.
While this implementation focused on reverse engineering of the COBOL codebase, it demonstrates how the legacy modernization components within AI/works™ can help lead developers and architects better understand legacy system design and complex codebases with minimal SME input.
