Augmented Reality: The Interface That Finally Grew Up

Augmented reality has had more false dawns than almost any consumer technology — Google Glass in 2013, the Pokémon GO summer of 2016, a decade of "next year" headsets. What changed recently is not a single product but the quiet maturation of the pieces: tracking that doesn't drift, displays that are bright enough outdoors, and AI models that understand what the camera is seeing. AR is becoming an interface rather than a novelty, and it's doing so first at work.

How AR actually works

Overlaying a digital object on the real world convincingly requires solving three problems continuously, dozens of times per second:

• Where am I? The device must know its position and orientation precisely. This is SLAM (simultaneous localisation and mapping): cameras and inertial sensors track visual features in the environment to build a map and locate the device within it. Depth sensors (LiDAR, structured light) help but modern systems do well with cameras alone.
• What's around me? Scene understanding — detecting planes, surfaces and objects — so that a virtual chair can sit on the real floor and be hidden behind the real sofa. Increasingly this uses neural networks running on the device.
• How do I show it? On a phone, the camera feed is composited with graphics — "video see-through". On glasses, light is projected into the eye through waveguides or similar optics while the real world is seen directly — "optical see-through". Headsets like Apple's Vision Pro and Meta's Quest use high-resolution cameras and screens to do video pass-through convincingly.

Each is hard on its own; doing all three on battery power in a package people will wear on their face is why consumer AR glasses took so long.

Enterprise found the use cases first

While consumer AR searched for a reason to exist, industry found several that pay for themselves:

• Remote assistance. A field technician wears a headset; an expert elsewhere sees what they see and draws annotations that stay locked to the equipment. Cuts travel and downtime; this was the first AR application with a straightforward ROI.
• Guided work. Step-by-step instructions for assembly, maintenance or wiring, displayed on the part itself. Boeing, Lockheed and others report large reductions in error rates and time on complex tasks.
• Logistics. Vision-picking in warehouses: the display shows which bin, the camera confirms the pick. Hands stay free.
• Surgery and medical training. Projecting imaging data onto the patient for navigation; practising procedures on virtual anatomy. Regulatory-cleared surgical AR systems exist for spinal and orthopaedic procedures.
• Design and construction. Walking through a building model on site, at scale, before it's built; checking installed work against the plan — closely tied to digital twins.

The consumer picture

Phone-based AR is ubiquitous and largely invisible: measuring apps, furniture previews, navigation arrows painted onto the street, face filters. The frontier is wearables, and it split into two paths. Spatial computing headsets — Vision Pro, Quest — offer immersive, high-fidelity experiences but are bulky and expensive, and adoption has been slower than the makers hoped. Smart glasses — Meta's Ray-Ban line and its successors with displays, plus a wave of competitors — went the other way: lightweight, camera and audio first, with an AI assistant that can describe what you're looking at, translate signs, or remember where you put your keys. These have sold in the millions, and the addition of small heads-up displays is the bridge toward true AR glasses.

AI is the missing piece

The reason smart glasses suddenly make sense is multimodal AI. A camera on your face is only useful if something understands what it sees. Models that can read a menu, identify a plant, summarise a whiteboard or give directions turn a wearable camera from a privacy liability into an assistant. Running enough of that on-device to protect privacy and cut latency is an edge AI problem, and the chips in current glasses are just crossing that threshold.

What's still hard

Battery life remains the binding constraint for all-day glasses. Optical see-through displays still struggle with brightness in sunlight and narrow fields of view. Social acceptance of always-on cameras is unresolved — bystander privacy is a real issue, not just a talking point. And the app ecosystem for spatial interfaces is immature: nobody has settled what the "window" or the "button" of AR looks like yet. The next few years are about those problems, not about whether the core technology works. It does.