AI Agents Start Spending Money: Stripe Sees a Jump, and Grok Promises Refunds
Stripe's Link says agent-driven purchases grew sharply in September, while Elon Musk reportedly promised to reimburse users if Grok mishandles a purchase. Both show where agent shopping is heading.
Two small stories point in the same direction: AI agents are beginning to spend real money, and companies are starting to say who pays when they get it wrong.
First, Stripe. Stripe is the payments company behind many online checkouts, and Link is its wallet product, with about 300 million users overall. Dan Hill, Stripe’s head of consumer product for Link, told Upstarts Media that roughly 200 agents now use Link to pay, and that agentic volume is “becoming very material”. An aggregator reports that agent transactions grew 38 times in September and payment volume 20 times. Those percentages sound huge because they start from a very small base. The detailed spending categories are behind a paywall, and the data comes from Stripe itself, so it cannot be independently checked. Also note it describes September, when about 70 agents were signed on, and the number has since more than doubled. Anthropic and OpenAI use Link, but not yet for agentic commerce.
Second, Grok. After a demo in which Grok reportedly booked a flight from a photo of a credit card, Elon Musk said SpaceX would reimburse users if Grok mishandles a purchase. He also said users would be covered if Grok leaked card information. Business Insider reported the promise, and the limits of the guarantee have not been specified, so it is a statement of intent, not a policy you can rely on yet.
What’s behind it: an agent that can pay is much more useful than one that can only suggest. But money raises the stakes on every earlier worry: wrong item, wrong date, wrong card, or a clever scam that tricks the agent. That is why banks and shops are also locking agents out in some places (see our report from October 18), and why payment companies are racing to build trusted paths for them. Reimbursement promises are one way to build trust, written terms are better.
What this means for you: if you try agent shopping, start small. Use a card with a low limit or a virtual card, require your approval before any payment, and check order confirmations. Before trusting a “we will reimburse you” promise, ask where it is written down. For most people, letting agents buy things unsupervised is still a bit early, and a human click at checkout is a cheap safeguard.
Sources
Source: https://www.upstartsmedia.com/p/data-stripe-link-how-ai-assistants-spend
Taste-Bench: AI Agents Are Surprisingly Bad at Choosing the Better Path
A new benchmark from Microsoft and City University of Hong Kong tests whether models can pick the better direction in a long task. The best score is under 60 percent.
