Agents In UseIn the next office-shop phase, Claudius gained browser access, CRM and better inventory context. It checked supplier prices and delivery details while human purchase approval remained mandatory.
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
For [request], inspect approved supplier pages and produce a quote draft with stock, total cost, delivery estimate and evidence links. Record unknowns in the CRM draft. Stop before purchase or customer commitment and ask for approval of the specific supplier and price.
Anthropic reports improved business performance but continuing manipulation and support needs.
Reported by Anthropic and Andon LabsResearchers withheld payment-interface access to require human purchase checks; humans stocked shelves and helped resolve problems.
A person approved each step before the agent continued.
Claudius carried forward product research and shop decisions, with new tools for ongoing state and follow-up.
Primary follow-up research report. Improvement is attributed to the authors, and robustness is explicitly limited.
Credit: Anthropic and Andon Labs
Source published 18 Dec 2025
Source checked 14 Oct 2026
Claude Sonnet 4.0 / 4.5 · Browser tools · Web search · CRM · Inventory tools · Slack · Google Forms · Payment-link tool
Operations · Checkpoints
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.
anthropic.comAnthropic and Andon Labs let a Claude-based agent manage a real office shop. It searched for products, set prices and requested restocking, but the researchers report losses and persistent mistakes.
Added
Claudius