Letting an agent go shopping sounds simple — until you think about credential handling and overspending protection. LangChain teamed up with Stripe to show how it's actually done. Title: Agents that can pay: building Restock with Stripe's Link and Managed Deep Agents URL: 📝 Overview LangChain built "Restock," an agent that autonomously orders real products by combining Stripe Link, the Machine Payments Protocol (MPP), the Zinc retail API, and its own Managed Deep Agents hosting platform. ❗ Problem it solves Browser-automation-based payment is fragile, exposing credentials to the model is dangerous, reversing bad transactions is hard, and agents need real guardrails against overspending. 🛠️ Method The agent searches and builds a cart via Zinc's product API, the user sets a spending ceiling, and a Slack-based human review (interrupt) checks the cart before any payment. When the merchant responds with 402 Payment Required, payment runs through MPP — but the actual authorization happens in a separate Link wallet interface the user controls, so the model never sees raw credentials. A tool verifies the final order matches the reviewed cart before submitting payment. 📊 Results In a real example — ordering a 12-pack of blue ink pens under a $25 budget — the final charge, including product, tax, shipping, and Zinc's fee, came to $22.18, with $0.82 automatically refunded. #AIAgents# #Payments#