# How did Rakuten speed up feature delivery with a coding agent?

> Rakuten uses Anthropic's Claude Code, an agentic coding tool, across engineering; in one test it worked autonomously for seven hours on a complex vLLM implementation and achieved 99.9% numerical accuracy versus the reference method. Anthropic reports Rakuten's average time to market for new features fell from 24 working days to 5 (79%).

- **Evidence grade:** B — Named organization with a quantified outcome, from the vendor or the organization's own announcement.
- **Organization:** Rakuten (technology company)
- **Industry:** Software, Internet & Media — E-commerce & internet services
- **Roles:** Software Engineering
- **Tools:** Claude Code (Anthropic)
- **Status:** production · **Agent pattern:** autonomous-agent · **Autonomy:** 3/4
- **Last verified:** 2026-10-17

## Reported outcomes
- time to market: 24 working days to 5 days (79% reduction) [1] (self-reported)
- autonomous run accuracy: 99.9% numerical accuracy vs reference method [1] (self-reported)

## How it works
1. Engineer assigns a complex task in natural language.
2. Claude Code works through a large codebase (vLLM, 12.5M lines) with occasional guidance.
3. Output is validated against a reference implementation.

## Risks & compliance
- Code review and testing remain required

## Notes
- Source page is undated; accessed 2026-10-17.

## Sources
1. Anthropic, "Rakuten accelerates development with Claude Code" (undated; accessed 2026-10-17) <https://claude.com/customers/rakuten>
   > "Engineering teams saw their average time to market for new features drop from 24 working days to 5 days—a 79% reduction that changed how Rakuten delivers innovation."

Canonical: https://agentcasefile.com/cases/rakuten-claude-code-engineering/ · License: CC BY 4.0 (summary) · Source: Agent Casefile