Fujitsu Application Transform: AI Agents Eat COBOL For Breakfast
Legacy modernization is the trillion-dollar problem everyone talks about but nobody solves well. Fujitsu just launched Application Transform powered by Kozuchi today, and the numbers are hard to ignore: 97% reduction in design document generation time, 95% more comprehensive output than generic AI, and 60% better readability.
The product analyzes COBOL and other legacy source code, then automatically generates design documentation that humans can actually read and act on. It uses a proprietary RAG architecture called Knowledge Graph-Enhanced RAG for Software Engineering that manages code analysis while preventing the hallucinations and omissions that make generic LLMs dangerous for this kind of work.
Why does this matter for the agentic ecosystem? Because legacy systems are the biggest untapped surface area for AI agents. There are billions of lines of COBOL still running banking, insurance, and government systems. Before agents can modernize that code, someone needs to understand what it does. Fujitsu's approach treats documentation generation as the first autonomous step in a multi-agent modernization pipeline, with code rebuilding, automatic rewriting, and maintenance support planned for later this year.
SMBC Nikko Securities has been validating the technology since fiscal year 2025. The product launches today as a SaaS in Japan, with introduction support services. This is Fujitsu, a $30B+ company, betting that AI agents can crack the legacy code problem that has resisted automation for decades.
https://global.fujitsu/en-global/pr/news/2026/03/30-01
← Back to all articles
The product analyzes COBOL and other legacy source code, then automatically generates design documentation that humans can actually read and act on. It uses a proprietary RAG architecture called Knowledge Graph-Enhanced RAG for Software Engineering that manages code analysis while preventing the hallucinations and omissions that make generic LLMs dangerous for this kind of work.
Why does this matter for the agentic ecosystem? Because legacy systems are the biggest untapped surface area for AI agents. There are billions of lines of COBOL still running banking, insurance, and government systems. Before agents can modernize that code, someone needs to understand what it does. Fujitsu's approach treats documentation generation as the first autonomous step in a multi-agent modernization pipeline, with code rebuilding, automatic rewriting, and maintenance support planned for later this year.
SMBC Nikko Securities has been validating the technology since fiscal year 2025. The product launches today as a SaaS in Japan, with introduction support services. This is Fujitsu, a $30B+ company, betting that AI agents can crack the legacy code problem that has resisted automation for decades.
https://global.fujitsu/en-global/pr/news/2026/03/30-01
Comments