How many people does an automated checkout quietly employ to fix what it cannot do alone? That question is why a three-year-old story about Amazon Just Walk Out is back on LinkedIn this week: a post from Boris Verbitsky pulled in 264 reactions and 252 comments retelling reporting that first broke in April 2024.

Automation removes what the customer sees

You automate the line, the register, the card swipe. That is what the demo shows you. Whatever the cameras cannot resolve on their own does not vanish. It moves to someone reviewing the exception on a screen, usually somewhere labor costs less than wherever you are selling.

Before you sign with any AI automation vendor, four questions matter more than the pitch about cameras and deep learning. What is the real exception rate? Who resolves it? Where does that person or team sit? What does it cost per hour? A vendor who answers those with confidence is a good sign. One who hesitates just showed you the margin they left out of the proposal.

What actually happened with Just Walk Out

Amazon Go, the first store running Just Walk Out, opened as an internal beta for employees on December 5, 2016, in Seattle. It did not open to the public until January 22, 2018, more than a year later, per Wikipedia. Not 2016, as the post making the rounds claims.

Amazon sold Just Walk Out as computer vision, sensor fusion, and deep learning, no line and no register. In April 2024, reporting tied to The Information found that more than 1,000 people in India were labeling video and reviewing the transactions cameras could not resolve on their own, according to Retail Technology Innovation Hub.

Amazon pushed back on its own site, calling the reports “erroneous” and stating that associates “don’t watch live video of shoppers to generate receipts, that’s taken care of automatically by the computer vision algorithms.” It said the human role was labeling and annotation after the fact, “no different than any other AI system that places a high value on accuracy, where human reviewers are common,” per its official statement.

Look at what Amazon denied and what it confirmed. It denied live watching. It confirmed a human team correcting what the model could not resolve alone, as a normal part of the system. The cameras were real. The machine learning was real. What the marketing left out was that team.

That same April, Amazon announced it was pulling Just Walk Out from its Amazon Fresh grocery stores and replacing it with Dash Carts, where shoppers scan their own items with a camera mounted on the cart, per Digital Commerce 360. The technology didn’t disappear. Amazon kept it running in Amazon Go stores and still licenses it to other retailers. But in a grocery format, where the exception count climbs with every new item on the shelf, Amazon decided it wasn’t worth the cost.

The small operator becomes the fallback

In the comments under Verbitsky’s post, someone identified as Abdullah M. brought the case down to the scale of a business that doesn’t have 1,000 people in India on standby. If you don’t, you are the fallback. He wrote a line worth keeping: “The tell shows up in the support inbox before the P&L does.” His test: if by month three your manual replies to edge cases outnumber the automated ones, the system was never actually finished.

That test scales down to any size company. Don’t just ask whether the system works. Ask how many times a week someone on your team, or you yourself, ends up fixing by hand what the AI was supposed to handle alone.

The map that asks that question before you buy

I already wrote about how AI moves the bottleneck instead of removing it. Just Walk Out is that pattern with a name attached: the bottleneck didn’t disappear, it moved to a room in India nobody in Seattle could see.

That is why the first phase of AI Maestro doesn’t start by picking a model. It starts with the Process Reality Map, which lays out where every exception actually lives today, who resolves it, how long it takes, and what breaks if that person calls in sick tomorrow. That map is what would have told you, before you signed with any “zero friction” vendor, how many real people were standing behind the camera.

Next time a vendor shows you a flawless demo, ask about the back room. If they don’t have an answer with numbers in it, you’re the one who is going to eat the exception.