Skip to main content
AI retailcomputer visionSilpo

Silpo AI Carts: Hype vs. Reality

At Silpo, there's no fully automatic deduction of money just by walking out, but there is a real AI cart pilot in Kyiv: items are recognized by cameras, and payment must be confirmed via smartphone. For retail, this is an important example of offline AI integration, where computer vision already reduces checkout friction.

Technical Context

I specifically checked what's happening at Silpo not based on rumors but on facts. The story about "you leave the store and the money is automatically deducted" sounds great, but as of July 2026, that's not the case: they don't have a full just-walk-out system like Amazon Go.

Instead, there's something that genuinely interests me as an AI implementation engineer. In Kyiv, Silpo is piloting AI-powered carts: the cart has cameras or a built-in scanner, the system recognizes the item when it's added, builds a virtual basket, and then I confirm the list and pay via smartphone.

This is an important difference. There is recognition, computer vision in physical retail exists, personalized prices through the app logically fit into this scenario, but there's no fully automatic charge upon exit yet.

In parallel, they've long had "Vilnokasa" (Free Checkout). That's a completely different mechanic: I scan barcodes myself in the app, then generate a QR code and go to a self-checkout. So it's not an OCR cart and not a cashierless store, but a neat Scan&Go.

And here, I wouldn't devalue the pilot. For offline retail, even this intermediate layer of AI integration is already useful: the system learns from real products, complex packaging, glare, overlaps, user errors, and integration with the loyalty program. On paper it looks simple, but in a store everything breaks against reality within half a day.

Business Impact and Automation

For retail, I see three practical effects here. First: less friction on the path to payment, especially if you're buying a few items and don't want to stand in line at the register. Second: higher app value, because loyalty, personalized discounts, and payment come together in one flow. Third: the store gets real data for further AI automation, not just a pitch for investors.

Who wins? Chains with high traffic and expensive checkout operations. Who loses? Those who try to jump straight to a "cashierless store" without proper AI architecture, data quality, and error control.

At Nahornyi AI Lab, I especially love such stories: not "let's slap a neural network on it," but let's build a working loop from CV to payment, discounts, and edge cases. If you have a retail or offline operation and want to build AI automation without fairy tales and with sound economics, we can easily discuss how to land it on your process.

We previously discussed how a lack of well-thought-out architecture turns AI demos into myths, using the example of the Codex 5.2 robot on Raspberry Pi. The success of Silpo with OCR carts demonstrates that smart integration turns a concept into a working solution.

Share this article