JapanTechnology and telecom2026 – ongoing

Rakuten

Specialist agents with memory built on Claude Managed Agents for business functions

Rakuten's pilot of memory-enabled Claude agents cut initial critical errors by 97%

In progressCompany's own claimVendor ↗

Problem

Rakuten wanted staff across engineering, product, sales, marketing and finance to use AI agents for their own work. Agents that forgot past mistakes kept repeating them.

Approach

Rakuten used Claude Managed Agents to run specialist agents in sandboxed environments, linked to Slack, Microsoft Teams and a task board. Long-running agents received goals rather than step-by-step tasks, and memory let them recall what went wrong in earlier sessions. The company tested the memory approach in a pilot.

Results

97%reduction in initial critical errors in the pilot, as stated by Rakuten's General Manager of AI for BusinessCompany's own claim[1]

more than 30%reduction in cost and latency in the pilot, with no loss in output quality, as stated by the same managerCompany's own claim[1]

Takeaway

Giving agents memory of past failures was the change Rakuten's AI lead credited for fewer errors and lower cost in the pilot. The page gives neither the pilot's size nor its duration.

Sources

Inside Rakuten's plan to turn every employee into a builder with Claude Managed Agents
Anthropic8 Apr 2026Vendor publication

Source summaries

[1]Anthropic8 Apr 2026

Anthropic's customer story describes Rakuten's company-wide push to put specialist Claude agents to work across business functions. It quotes Rakuten's AI lead on a pilot with memory-enabled agents that cut initial critical errors by 97% and cost and latency by more than 30%. The page carries no date; the Internet Archive first saved it on April 8, 2026.

Full details are in the original sources.

Added 4 Nov 2026 · Updated 4 Nov 2026