Sony Bank Cuts Core Banking Development Time 30% Using AI Coding Agents
Sony Bank and Fujitsu have run Anthropic's Claude and Claude Code on Amazon Bedrock across design, coding and testing of Sony Bank's core system since last September. The companies report a 30% cut in development time and a 40% cut in labor hours, with human engineers still reviewing the output.
TOKYO — Sony Bank and Fujitsu said they have cut development time on Sony Bank's core banking system by 30% and labor hours by 40% since bringing generative AI coding agents into live production work roughly a year ago. The companies have been running Anthropic's Claude and Claude Code models, accessed through Amazon Bedrock, across design, coding and testing on Fujitsu's Core Banking xBank platform, which runs on AWS infrastructure.
The engagement began in September 2025 and the companies disclosed the cumulative results as of July 2026 — roughly ten months of production use, not a pilot-stage estimate. That distinction matters in an industry where AI-assisted software development claims are common but rarely accompanied by a specific, dated before-and-after figure tied to a named production system.
Core banking systems are among the most conservative software environments in financial services: they process account balances, transaction ledgers and settlement logic where a coding error can directly corrupt customer money, and banks have historically been slow to let any automated tool touch that code without extensive review. Sony Bank and Fujitsu's description of the workflow keeps human engineers in the loop at each stage — the AI agents assist with design, manufacturing and testing, but the companies say human oversight remains central to catching quality and safety issues before code ships.
The two companies say they now intend to extend AI assistance further upstream and downstream in the development lifecycle, into requirements definition — the stage where business needs are translated into technical specifications — and into ongoing system maintenance, which for a bank's core platform is typically a multi-year, resource-intensive obligation that continues long after initial development ends.
Fujitsu has positioned Core Banking xBank as a cloud-native alternative to the mainframe-based systems that still run much of retail banking in Japan and other markets, and the generative-AI development story is as much a pitch to other banks considering a similar migration as it is a Sony Bank efficiency case study. Whether the reported time and labor savings generalize to other institutions' codebases, or reflect efficiencies specific to Sony Bank's particular migration, is not something the companies' disclosure addresses.