Rakuten, a Japanese technology company with over 70 businesses spanning e-commerce, travel, fintech, digital content, and communications, deployed Claude Code to compress time-to-market for new features from 24 working days to 5 days — a 79% reduction — while sustaining 7 hours of autonomous coding on a complex open-source refactor with 99.9% numerical accuracy.
"We want to give all our teams the power to innovate quickly and drive greater impact for customers faster. It's about multiplying what each team can achieve, not just automating existing tasks," said Yusuke Kaji, General Manager of AI for Business at Rakuten.
The pressure: thousands of developers serving millions of customers across Rakuten's 70+ business units — and previous AI coding tools required constant human guidance, couldn't navigate complex multi-language codebases, and didn't scale. "Speed and ROI are our key metrics. We measure success by how quickly we can deliver value to customers, not just how fast AI can write code," Kaji explained. Rakuten's "AI-nization" strategy already built in-house agents and models, giving the team intuition about where AI assistance fits.
Why Claude Code: terminal-native, handles multi-language enterprise codebases, and aligns with Anthropic's safety commitments. When Anthropic released Claude Managed Agents, Rakuten shipped specialist agents across product, sales, marketing, and finance — each deployed within one week, plugging into Slack and Teams. "With Managed Agents, our power users become like Galileo, contributing across domains far beyond a single specialty or discipline," Kaji said. The breakthrough: machine learning engineer Kenta Naruse asked Claude Code to implement an activation vector extraction method in vLLM (12.5 million lines, multiple languages). "I didn't write any code during those seven hours. I just provided occasional guidance," Naruse recalled.
Architecture: Claude Code (terminal-native, edits files, runs commands) integrates into developer workflows — unit tests, API mocking, components, bug fixes, docs, plus AI-powered code review for pull requests and parallel sessions to eliminate bottlenecks.
Operational scale: engineering teams ship features in days instead of weeks; non-engineers contribute via the terminal interface with guardrails; test coverage and "AI-nization ratio" tracked alongside speed. Senior ML engineer Diego Mateos says "It generates comprehensive tests instantly, then builds features that pass them. It's completely changed how I develop." Manager Manoj Desai: "Claude Code gives us those super powers to make executions much, much faster."
Future direction: Naruse is building an "ambient agent" that splits complex monorepo updates into 24 parallel Claude Code sessions — work that would take over a month manually. Rakuten plans to scale Claude Code from quick tasks to multi-hour workflows across the developer organization, pursuing an enterprise where technical barriers no longer limit innovation.