I built a Claude Code skill that lets me buy coffee beans without leaving my terminal. You type /coffee-me, it searches Shopify’s Global Catalog, shows you options with prices and ratings, you pick one, and it opens checkout in your browser. The whole thing takes about fifteen seconds. I wrote it as an experiment with Shopify’s new agent APIs, and the strangest part isn’t that it works, it’s that it feels so much nicer than trawling Google search. You’re sitting there writing code, you’re out of coffee, an AI finds you some, and you’re back to work. No multiple tabs, no browsing, no comparing seventeen Ethiopian single-origins. Just pick and you go.
The agentic commerce hype machine projects somewhere between $190 and $385 billion in AI-driven shopping by 2030. But today, only 14% of Americans say they’d trust an AI to place an order for them. There’s a nice tension there. The bull case assumes people will let agents handle complex purchases, flights, furniture, electronics, but complex purchases are exactly where trust matters most. And simple purchases? Buying coffee beans from the terminal while you’re deep in a coding session? That might actually be the right stakes for us to start with. Low enough that you don’t care too much, high enough that there’s a decision involved.