Fujitsu has released a generative AI service designed to analyse source code and automatically generate design documents. The tool, named Fujitsu Application Transform powered by Fujitsu Kozuchi, aims to assist organisations in modernising legacy systems by significantly reducing the manual labour involved in system analysis. Starting March 30, 2026, Fujitsu began offering this service as a Software as a Service (SaaS) model in Japan.
The service targets COBOL and other legacy programming languages commonly found in older corporate systems. According to Fujitsu, the AI reduces the time required to understand code and produce documentation by approximately 97%. This allows teams to grasp system specifications without needing specialised knowledge of outdated languages.
Key performance metrics provided by the company include:
- 95% Improvement in Comprehensiveness: Compared to standard generative AI, Fujitsu’s system captures consistent design information without missing complex segments in COBOL.
- 60% Better Readability: The generated documents are more structured and easier to follow than those produced by conventional manual or automated methods.
- Accuracy Measures: The service utilises Fujitsu Knowledge Graph-Enhanced RAG (Retrieval-Augmented Generation). By linking vast amounts of source code through a knowledge graph, the system minimises “hallucinations”, a common issue where AI generates false information, and ensures no critical code logic is omitted.
SMBC Nikko Securities Inc. has already engaged in joint verification of this technology. Toshihiro Horiuchi, the firm’s Managing Executive Officer, stated that the company has been testing the reverse-engineering of COBOL design documents since 2025. He noted that the initiative provides a realistic path for updating legacy systems by combining AI with long-standing development expertise.
Roadmap ahead
Fujitsu plans to expand the service’s capabilities throughout fiscal year 2026. Some of the upcoming features include automatic source code rewriting, system rebuilding, and maintenance support. And, by addressing the ‘hardware-software gap,” where modern digital goals are slowed down by ancient underlying code, these tools aim to keep organisations competitive without starting from zero.