The Cashier and AI: The 'AI' Store Had 1,000 People Watching the Video

In 2022, roughly 700 of every 1,000 transactions in Amazon's "Just Walk Out" stores, marketed as computer vision so advanced customers could pick up items and leave without a checkout at all, were actually verified by workers in India reviewing camera footage. The system wasn't fake. But it was far more human-in-the-loop than the "AI" branding suggested, and by Amazon's own account it had become too slow and too expensive to run at scale. In April 2024, Amazon pulled Just Walk Out from most of its US Amazon Fresh grocery stores and replaced it with a smart shopping cart instead.

Cashiers are rated High risk, with a horizon toward 2027, the most urgent classification this series applies. That risk is real and well underway. But the Just Walk Out story, and what happened to self-checkout more broadly over the following year, is a useful corrective to the assumption that "automated retail" means what the marketing says it means.

Self-checkout itself, the older and far more widespread technology, has been quietly reversing in places too. Dollar General pulled self-checkout entirely from 300 of its highest-theft stores in 2024. Booths, a UK grocery chain, removed it from its stores altogether, telling reporters simply that customers wanted to talk to people. Neither of those decisions were about the technology failing to scan items correctly. They were about what happens to a store once nobody's watching the register.

What a Cashier Actually Does

The core task, scanning items and processing payment, is genuinely simple and has been automatable in principle for years. What surrounds it is less so: verifying age for alcohol and tobacco purchases, handling items without barcodes, catching honest mistakes and dishonest ones, and providing the kind of visible presence that functions as loss prevention whether or not anyone frames it that way. Most of the friction in retail automation over the past two years has come from that surrounding layer, not the scanning itself.

The Just Walk Out Story

Just Walk Out was Amazon's most ambitious cashier-replacement bet: ceiling-mounted cameras and computer vision tracking what a shopper picked up, charging their account automatically on exit. Reporting from Bloomberg, the Washington Times, and Business Standard in April 2024 revealed that a substantial share of that "computer vision" was, in practice, backstopped by roughly 1,000 workers in India manually reviewing footage to confirm what the algorithm had flagged. Amazon maintained the workers were checking a small minority of transactions for accuracy rather than doing the core work. Either way, the system proved too slow and too costly to run profitably at grocery-store scale, and Amazon removed it from most of its US Fresh stores that same month, shifting toward Dash Cart, a smart shopping cart that tracks items as a customer bags them rather than replacing the checkout moment entirely.

The Self-Checkout Backlash

Self-checkout's problems are better documented and more purely economic. Research from computer-vision company Grabango found self-checkout lanes running a shrink rate around 3.5% of sales, more than sixteen times the roughly 0.21% rate at staffed registers, with the National Retail Federation tracking overall retail shrink climbing toward $132 billion by 2024. Dollar General responded in 2024 by removing self-checkout from 300 of its highest-shrink locations and converting roughly 9,000 more to an "assisted" model with strict item limits, after its CEO called shrink the company's most significant headwind. Walmart and Target both scaled back self-checkout availability the same year, and Booths in the UK removed it from its stores entirely, saying plainly that its customers preferred talking to a person.

What's Actually Replacing Cashiers

The technology that is spreading isn't full autonomy. It's smarter carts and better-instrumented self-checkout. Amazon's Dash Cart and Instacart's Caper Cart, now deployed across thousands of independent and regional grocers, use onboard cameras and weight sensors to track items as customers shop, letting them skip the checkout line entirely while keeping the transaction fully self-directed and auditable. AiFi's computer-vision checkout now runs in roughly 20 sports stadiums. Grabango, whose underlying technology reportedly worked well at Aldi's Aurora, Illinois pilot, closed entirely in October 2024, a reminder that working tech and a viable business are not the same claim.

What Customers Actually Think

A Drexel University study published in 2024 found that staffed checkout was associated with higher customer loyalty than self-checkout, an effect that grew stronger with basket size, customers with larger purchases responded better to a person than a machine. A CNN-cited survey found 67% of shoppers had experienced a self-checkout failure, with a quarter saying they were more likely to avoid it afterward. At the same time, industry survey data puts self-checkout preference for speed at 77% among shoppers, with 43% still wanting a staffed attendant available nearby. The honest summary is that customers want the machine for convenience and the person for reassurance, and no retailer has found a way to deliver both without keeping some humans on the floor.

The Labor Market Reality

The U.S. Bureau of Labor Statistics projects cashier employment declining 6% from 2025 to 2035, a loss of roughly 200,600 positions from a base near 3.1 million, with a median wage of $15.81 an hour. The agency names self-service checkout and online sales growth directly as the drivers. Unlike several professions in this series, there is no credible case that this decline reverses. The technology replacing the transactional core of the job, however imperfectly branded, is real, already deployed, and getting cheaper.

How to Use AI as a Cashier Now

For the transactional core: assume it keeps shrinking. Self-checkout and smart-cart technology handle the scan-and-pay function well enough that resisting the shift is not a viable long-term position.

For the surrounding work: age verification, exception handling, and visible loss-prevention presence are exactly the tasks retailers have found they still need a person for, evidenced directly by Dollar General's and Booths' reversals. Building skill and reliability in that layer, rather than pure transaction speed, is where the remaining demand concentrates.

For career direction: retail roles shifting toward customer service, loss prevention, and store operations carry more durable demand than pure checkout work. The BLS decline is specific to the transactional role, not to retail employment broadly.

What I Think

The 2027 horizon and High-risk rating are, on the evidence, hard to argue with, and the Just Walk Out story doesn't really undercut that, it refines what "automation" actually looks like on the ground. The lesson isn't that AI failed to replace cashiers. It's that the version of AI expensive enough to need human review at scale lost to a cheaper, more honestly labeled alternative, a smart cart, that does less and works better.

What I'd take most seriously here is the shrink data, because it reframes the self-checkout retreat as an economic decision, not a technological one. Retailers didn't rediscover a preference for human warmth. They discovered that removing a person from the register costs more in theft than it saves in labor, in enough stores that several major chains reversed course within the same year. That is a more durable protection for this job's surviving edges than customer sentiment alone would provide.

"The store that promised to replace the cashier needed a thousand people watching the video to work. The technology that actually stuck was the one that admitted it still needed a person nearby."