The narrative that got it backwards
When Amazon shut down its Go and Fresh cashierless grocery storefronts, the easy read was that autonomous checkout had failed as a category, an expensive experiment that consumers never fully embraced for a full weekly shop. That narrative missed what Amazon was doing at the same time in a much less visible part of its business: expanding Just Walk Out technology to more than 40 North American fulfillment centers, where employees grab items without stopping at a register, and licensing the same underlying system to stadiums, airports, hospitals, and campuses well beyond its own retail footprint.
That is not a company retreating from a failed bet. It is a company that learned where the technology actually earns its cost and redeployed accordingly, moving the same camera and sensor stack into settings where the economics were always going to work better than a full-size supermarket. The distinction matters for every retailer that watched the Go and Fresh closures and concluded autonomous checkout was a dead end, when the more accurate read is a category still finding its real market and now finding it at genuine scale, across more locations than the original consumer-facing storefronts ever reached.
Speed environments, not shopping environments
The pattern across Just Walk Out's surviving and expanding deployments is consistent: fulfillment centers, stadiums, airports, hospitals, transit hubs, and campus breakrooms. Every one of those environments shares a trait that a full grocery store does not, the shopper wants to grab a small number of items and leave as fast as physically possible, with zero tolerance for a line or a scanning delay slowing down a shift break or a flight connection.
That is a fundamentally different design target than a weekly grocery run, where shoppers browse, compare, and fill a cart over twenty or thirty minutes, and where a cashier or self-checkout lane was never really the bottleneck in the shopping experience to begin with. Autonomous checkout's value proposition is strongest exactly where transaction speed is the primary thing being purchased, and weakest where it offers only a marginal improvement on an already-tolerable wait at a register most shoppers barely notice.
Retrofit is beating rebuild across the vendor field
Reckon.ai, one of the vendors competing alongside AiFi, Sensei, Trigo, and Zippin in this space, describes its approach as converting existing cabinets, fridges, and shelves into autonomous assets rather than requiring a retailer to build a purpose-designed store from the ground up. That retrofit-first posture mirrors what is happening in the loss prevention side of computer vision retail tech as well, where vendors gaining traction are the ones that bolt onto existing infrastructure instead of demanding a wholesale rebuild before value shows up.
Reckon.ai frames the broader shift plainly: while Amazon changed course in large grocery stores, it doubled down elsewhere, and the idea evolved rather than collapsed when the pioneer changed course. That evolution is measurable in deployment count rather than press coverage, hundreds of third-party locations now running the same underlying Just Walk Out stack that once powered a handful of flagship Amazon storefronts built explicitly to prove the concept to the public before the company understood which environments would actually pay for it at scale.
What actually failed, and what did not
The honest accounting of Amazon Go and Fresh is that the format failed commercially, the cost of instrumenting a full-size grocery store with enough cameras and weight sensors to track every item never penciled out against the format's actual traffic and basket size. That is a real failure, and it should inform how any retailer models the economics of applying this technology to a large-format store rather than the smaller, faster environments where it has since proven itself out.
The computer vision and sensor fusion technology underneath the format kept improving and kept finding paying customers in environments where the unit economics work far better: high-frequency, low-basket-size, time-pressured transactions where every second removed from the checkout process is worth real money to the operator. Conflating the format failure with a technology failure would lead a retail technology leader to the wrong conclusion at exactly the moment the underlying stack has become mature enough to deploy with real confidence and a growing base of reference customers.
The evaluation framework this suggests
For a retail CTO evaluating autonomous checkout today, the Amazon experience argues for a narrower and more disciplined first question than "should we go cashierless." The better question is which specific formats in the existing store or facility portfolio, a staff break room, an airport concourse kiosk, a hospital cafeteria, a stadium concession stand, actually match the profile where this technology has proven itself: high transaction frequency, small basket size, and a customer who values speed above almost everything else, including price comparison or brand exploration that a slower channel still supports well.
Starting there, with a retrofit approach bolted onto existing fixtures, produces both a cheaper pilot and a return worth scaling from day one. The vendors now winning deals, Amazon's own Just Walk Out licensing business alongside the AiFi, Sensei, Trigo, Zippin, and Reckon.ai field around it, are the proof that this has become genuine infrastructure with a track record, deployed at a scale that a novelty product still hunting for its first real use case could never reach.
