OpenAI Wants ChatGPT to Buy Things for You. Should We Be Worried?

But demos always look slick. I've sat through hundreds of them.

Key things we don't actually know yet:

• How many merchants are actually integrated (OpenAI hasn't disclosed figures)
• What happens when the AI makes a mistake, and who's liable
• Whether this works for anything beyond simple consumer goods
• How it handles returns, disputes, or the AI ordering the wrong size
• What data merchants get about your preferences and browsing

The business-focused updates they announced in March suggest they're thinking about enterprise applications too. Which makes me wonder if the real play here isn't consumer shopping at all, but procurement. Getting ChatGPT to reorder office supplies, manage vendor relationships, that sort of thing.

Now that would be interesting. When I was at Kuka, our procurement process was a nightmare. Half the delays on the factory floor came from waiting for parts that someone forgot to order. If an AI could actually manage that reliably, well, I'd pay attention.

But consumer shopping? I'm less convinced. The shopping research feature seems genuinely useful for comparison shopping. I tested it myself, asked about cordless drills. The recommendations were reasonable, though it missed some brands I know are solid (DeWalt got short shrift, which my old colleague at Bosch would find amusing).

The instant checkout stuff feels premature. It's too early to say whether people will actually trust an AI to complete purchases for them. Maybe younger folks will. My daughter lets her phone do all sorts of things I wouldn't dream of. But there's a difference between "here are some options" and "I just spent your money."

I called my old colleague at Siemens last week, we still chat occasionally. He's been following the AI commerce space. His take: the technology is probably ready, but the liability frameworks aren't. Who's responsible when the AI buys you the wrong thing? When it gets scammed by a fake merchant? When it misunderstands your request and orders 50 units instead of 5?

These aren't hypotheticals. Anyone who's worked in industrial automation knows that edge cases are where systems fail. And commerce has a lot of edges.

So am I worried? Sort of, but not for the reasons you might think. The technology will improve. It always does. What concerns me is the rush to deploy before the boring stuff (returns, disputes, fraud prevention) is figured out. OpenAI is moving fast because they have to. Competition is fierce. But moving fast in commerce means real money changing hands based on AI decisions.

We'll see. I've been wrong before. That shopping cart in Hanover was genuinely terrible, but now I see autonomous robots in warehouses every day. Maybe this is the same trajectory. Maybe in five years we'll wonder why we ever typed our credit card numbers manually.

Or maybe it'll be another demo that looked great on stage.

Robert "Bob" Macintosh · 1 hour ago · 3 min