So Amazon put out this video earlier in the week for a store called Amazon Go, and Ive probably watched it four times now, partly because Im impressed and partly because Im trying to spot where its faking it.
Quick version if you missed it: theres a physical convenience store, about 1,800 square feet, opening up in Seattle at 7th and Blanchard. You walk in, scan your phone at a turnstile-looking gate, grab whatever you want off the shelves, and just walk out. No register, no line, no swiping a card. The app figures out what you took and charges you automatically when you leave. Amazon is calling it "Just Walk Out Technology," which is a very Amazon thing to name something, half marketing and half literally just describing the feature.
The tech behind it is apparently a mix of computer vision, sensor fusion, and deep learning models watching the shelves, kind of like what they're building for self-driving cars but pointed at a bag of chips instead of a highway. When you pick something up it goes in your virtual cart, put it back and it comes out. In theory.
Heres my problem, and its not really with Amazon specifically, its with the entire concept of "the machine knows what youre doing" retail tech, because Ive been burned by this before, repeatedly, at my regular grocery store about ten minutes from my apartment.
You know the self-checkout lanes. Four registers, one bored employee standing guard over all of them holding a giant ring of override keys like a prison warden. I scan a bag of clementines, put it in the bagging area, and the little screen goes "UNEXPECTED ITEM IN BAGGING AREA" in red letters, red light spinning, like Ive committed a felony. I didnt do anything wrong. The scale is just too sensitive, or not sensitive enough, or the bag shifted, who knows. Now Im standing there with my hand up waiting for the warden to notice me over the beeping. This happens to me at that store legitimately one out of every three visits. I've started doing math in my head about whether self-checkout is actually faster than just waiting in the regular line, and some weeks Im honestly not sure it is.
So when I watch the Amazon Go video and see someone grab a sandwich off a shelf and just stroll out the door with zero friction, my first reaction isnt "wow, the future," its "okay but what happens when I put the sandwich back because I decided I wanted the other kind instead, and the camera loses track of it, and Im standing outside the store two blocks later getting charged for a sandwich I dont have." Maybe the system is genuinely better than a $20,000 grocery store scale. It probably has to be, honestly, because if it's not more reliable than that, nobody's going to trust it either.
The other thing nagging at me, and I say this as someone who does not usually go in for the "robots are taking our jobs" hand-wringing, is that this store apparently doesnt need cashiers at all. Not fewer cashiers, none. Theres still staff for stocking shelves and prepping food, from what I can tell in the video, but the actual point of a cashier, standing there scanning and bagging, just isnt a job in this building anymore. Im not saying dont build it. Im saying its worth sitting with for a second longer than most of the coverage this week has, which is mostly just "look how cool this is" and then moving on.
Right now its a single store in Seattle, opening to Amazon employees first sometime early next year as a kind of extended beta before regular people get let in. So none of us are actually testing this ourselves anytime soon, which is probably smart on Amazons part, let a few thousand employees find the edge cases before the general public does. Im curious whether it holds up once its not a controlled demo with camera-friendly lighting and people who know exactly how theyre supposed to move through the space.
If it works the way the video shows, fine, Ill eat my words and my sandwich. Ive just watched enough blinking red lights over unexpected items in bagging areas to want to see it running for six months with real chaotic customers before I get excited.