In brief: Claude Fable 5 enables Rakuten to run agents unsupervised for hours, as the model continuously verifies and corrects its own work during execution.
Rakuten uses Claude Fable 5 to build AI agents that run for days without human intervention while continuously checking their own work. This allows the company to accelerate its transformation into an AI-native enterprise.
Since March 2025, Rakuten has been building agents with Claude distributed across various business functions—including product, sales, marketing, and finance. These are integrated via Slack, Microsoft Teams, and Rakuten’s own task systems. Yusuke Kaji, General Manager of AI for Business at Rakuten, relies on so-called “stretch tasks” when working with new models to probe performance limits. Rakuten calls its company-wide AI initiative “AI-nization”—the goal is to integrate AI into all customer processes, business partnerships, and employee tasks.
The central challenge with long-running agents has been error tolerance: if an agent made a faulty decision in its initial steps, the problem compounded throughout the entire runtime. Earlier models did not detect these deviations during execution, only when a human reviewed the results—often hours later. For five-hour or multi-hour tasks, an early error could ruin the entire runtime. Rakuten had to continuously monitor manually.
Claude Fable 5 fundamentally changes this equation: the model repeatedly verifies its own intermediate results during runtime and corrects itself when it detects a wrong path. This is the first time Rakuten has seen agents handle autonomous multi-hour or multi-day tasks “overnight” without human intervention. Rakuten reports that its agents close issues roughly ten times faster across all areas than was previously possible.
Kaji sees this as a paradigm shift: “For the modern enterprise, it is about minimizing communication costs. Agents can equally demonstrate where they bring innovations faster into production—from idea to solution in the shortest time.” With agents that maintain context and map business logic, individual qualified employees could multiply their potential a hundredfold. However, the new bottleneck is not technical execution, but human judgment: as agents become faster, the number of tasks Rakuten takes on grows in parallel—without an increase in human evaluation expertise.
Source: claude.com · Published 19 July 2026
Lumi AI News — AI-assisted curation pursuant to Article 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.