Working at the frontier: How Rakuten builds agents overnight with Claude Fable 5
02:00 · July 20, 2026 · Claude Blog
Summary
Rakuten has deployed Claude Fable 5 to run autonomous agents across product, sales, marketing and finance functions. General Manager of AI Yusuke Kaji reports that the model sustains multi-hour and even overnight tasks with far less human intervention than earlier versions. Its improved self-reflection and self-verification allow it to detect and correct early errors before they compound, turning what used to be high-risk, multi-step jobs into reliable overnight runs.
This capability has changed how Rakuten structures work for agents. Teams no longer need to decompose every request into tightly scoped chunks; instead, they can hand over an entire business objective and let the agent maintain context, revisit assumptions, and carry memory of prior failures into subsequent sessions. The result is a shift in team effort from writing and debugging code to clarifying the underlying business problem and exercising judgment on outcomes.
Even with these gains, human oversight remains the binding constraint. Although agents now close issues roughly ten times faster, the volume of tasks continues to rise, so progress still depends on people to set priorities and validate high-stakes decisions. Kaji therefore routes only the work that benefits most from frontier-level reasoning to Fable 5, while routing simpler tasks to smaller models to control cost per completed task.
Looking ahead, Rakuten is exploring agents that coordinate across teams, preserving context and intent that would otherwise be lost in hand-offs between people. Kaji views the current model as one that has crossed a threshold in autonomy, yet notes that organizations are still learning the most effective ways to match such systems to real business workflows.
Why it matters
This article provides product teams and builders with insights into deploying long-running, autonomous AI agents using Claude Fable 5. It highlights practical enterprise strategies for balancing model intelligence with cost and managing human-in-the-loop constraints, which are highly applicable to Dutch AI practitioners scaling agentic workflows.
