<?xml version="1.0" encoding="UTF-8"?><oembed><type>video</type><version>1.0</version><html>&lt;iframe src=&quot;https://www.loom.com/embed/7d9e7c2100a1404fbc9be4f16a613b29&quot; frameborder=&quot;0&quot; width=&quot;1658&quot; height=&quot;1243&quot; webkitallowfullscreen mozallowfullscreen allowfullscreen&gt;&lt;/iframe&gt;</html><height>1243</height><width>1658</width><provider_name>Loom</provider_name><provider_url>https://www.loom.com</provider_url><thumbnail_height>1243</thumbnail_height><thumbnail_width>1658</thumbnail_width><thumbnail_url>https://cdn.loom.com/sessions/thumbnails/7d9e7c2100a1404fbc9be4f16a613b29-a8f19c7b751b54be.gif</thumbnail_url><duration>147.093</duration><title>ProofCard, AI Agent for Safe Buying</title><description>This Loom explains ProofCard, an AI agent that safely automates purchasing by reconstructing scattered Slack quote threads and ranking supplier offers. It uses DeepSeek to build current offers and excludes deals that are wrong or unreliable, such as Acme being recommended while Bolt is the cheapest at $878 but still excluded for missing delivery by the deadline, and Xerox having missing shipping so totals are not real. ProofCard only settles after user approval, verifies the owner’s authorized messages, rechecks that the deal has not changed, and then executes payment through an admin code tied to the negotiation with payment logs. The team reports 14 of 14 scenarios passing against a baseline with guardrails removed, with 0 unauthorized payments and exactly 1 charge under failure conditions, and concludes it can let the agent pay because it earns it.</description></oembed>