When something goes wrong with the appliances in your home, what do you do to fix them? Probably, you Google the problem. Then you dig out a manual or look one up online. If none of that helps, you might call the company who made the thing, and then spend an age on the phone trying to explain what’s gone wrong.

But what if you didn’t have to explain — what if you could just show someone the problem, and have it explained to you? That’s the proposition from Israeli company TechSee, which is building a customer support platform using two of 2017’s most overused buzzwords: augmented reality and artificial intelligence. But, unlike a lot of firms getting on the AR/AI bandwagon, TechSee’s proposal might actually be useful.

It works like this. When something goes wrong with product X (your router, for example) you call the company that made it. They send you a text with a link to a web app, which you tap to access TechSee’s platform on your browser. This then lets a customer service rep see a live view from your phone’s camera. The rep can highlight stuff on their display, and the markings appearing on your phone so you know to connect cable X, or flip switch Y.

Where things gets particularly clever (or scary, depending on your feelings about job automation) is that TechSee will then use these videos to train an AI program to do the troubleshooting in future. So instead of a human telling you to connect cable X or flip switch Y, these instructions will just pop up on your screen as automated prompts. By identifying the products and problems over AR, the human customer service reps are, essentially, training their robot replacements.

The reason this is clever is that it incorporates the tricky work of training an object recognition system without any extra effort. Regular, off-the-shelf vision services sold by Google and Amazon are great at identifying generic categories like bike, cat, or car, but they can’t can’t spot specific products, like D-Link’s DSL-6850U router. They were just never trained to. TechSee’s platform will be doing this training in the background, even as it provides its human-assisted AR platform; giving companies an on-ramp to automation without extra expense.

The company’s AR platform is already used by dozens of firms for tech support including Vodafone, Orange, and Samsung, and this services includes some basic object recognition that helps customer support agents identify problems and solutions. Speaking to The Verge, founder Eitan Cohen said TechSee was also building out its completely automated AI platform for two other companies, but that he couldn’t name them as the service was still in beta.

“The image recognition system is very unique, because it can identify a device down to the model number,” Cohen told The Verge. “It analyzes display errors, like codes that appear on screens [and] it recognizes different model just by looking at features — like button layouts, text size, colors, and so on.” It’s basically facial recognition for your appliances.

There are still a lot of potential problems with the system, especially if the AI fails to recognize what’s in front of it, or keeps on suggesting the wrong problem. But that’s nothing that can’t be fixed by having a human on hand to step in. TechSee is confident it’s on the right track too, and says initial feedback has been uniformly positive. “Customers are happier, call-time goes down, and fewer technicians need to dispatched,” says Cohen. “Everyone wants AI as the future of customer care.”