AI glasses: from a demo to a pilot that proves something

Somebody on the operations team has watched a launch video. A person in display glasses looks at a shelf, asks the assistant what is missing, and the answer appears in the corner of their view. The next question in the meeting is obvious: could our pickers, technicians or inspectors do that?
The glasses in question are real. XREAL Aura, announced for sale on 7 October 2026, is a pair of optical see-through glasses running Android XR, developed with Google, with Gemini as the built-in assistant. They weigh under 95 g, show a 70° field of view and connect by cable to a compute puck with a Qualcomm Snapdragon chip rated at 48 TOPS for AI work. A high-resolution camera with a privacy light gives the assistant a view of what the wearer sees. Prices start at USD 1,279, according to XREAL’s own product page.
None of that tells you whether the glasses will make one of your tasks faster, safer or cheaper. A demo shows what the device can do in a prepared room. A pilot has to show what it does in yours.
Why the demo answers the wrong question
A launch demo is built to show range: translation, navigation, recognising an object, opening an app by voice. Your operation does not need range. It needs one task done better than it is done today, by people who are also carrying things, wearing gloves and talking to colleagues.
The gap between the two is where hardware pilots usually go wrong.
The task is a theme, not a task. “Hands-free assistance in the warehouse” cannot be measured. “Confirming the right part at the third check of a returns inspection” can.
The data path is assumed. The agent sees through the camera, but the answer it needs — the bill of materials, the job sheet, the location of the part — lives in your systems. Until somebody decides how the agent reaches them, and with whose permissions, the demo is answering from general knowledge.
Nobody knows where the work runs. XREAL does not publicly describe which parts of the AI work happen on the puck and which in the cloud. That is not a criticism; it is a question you have to ask, because it decides what leaves the building and how the system behaves when the network drops.
The bystanders are forgotten. A camera worn at head height sees colleagues, customers and documents on other people’s desks. The privacy light helps the people around the wearer know it is recording. It does not decide what is kept, for how long, or who may watch it later.
Availability is not checked. XREAL ships first to the United States, Canada, Japan and South Korea, with further markets in 2027. Poland and the rest of the EU are not in the first wave. A European pilot planned for this quarter has no supported device to run on.
What a pilot that proves something contains
Infographic
Five things to write down before the first pair is unboxed
- One taskNamed in operational terms, done today by named people, with a known current time and error rate you can measure.
- The data it touchesWhat the camera and microphones capture, which company systems the agent reads, and what it may write back.
- The measurementThe same measure before and during the pilot, collected the same way, by somebody other than the supplier.
- The full costDevices, spares, enrolment, support time and the hours of the crew taking part, not only the hardware price.
- The stopping conditionThe result, incident or cost that ends the pilot early, agreed before it starts.
Scope it to one crew and one task. One returns bench, one maintenance round, one inspection line. Small enough that you can watch every shift for the first weeks, which is how a pilot with one crew becomes a specification for everything after it.
Map the data before the device arrives. List what is captured, where it goes, how long it is kept and who can see it. Ask the device and platform vendors the questions their documentation does not answer, in writing. Unanswered questions belong on the risk register, not in a footnote.
Measure the task, not the enthusiasm. Time per item, rework, misses, near misses. Satisfaction surveys from the first week of a new gadget measure novelty. Keep the measurement going after the novelty wears off; that is when the real answer appears.
Cost the estate, not the purchase. Glasses with a cabled puck add charging, cleaning, prescription inserts or fitting, spares and enrolment to the usual device questions. We wrote about the device line that is missing from most business cases; it applies with more force to something worn on the face.
Write the stopping condition first. Agree in advance what would end the pilot: no measurable change after a set number of shifts, an incident involving recorded bystanders, a support load the team cannot carry. Writing it down before the pilot starts is what keeps the decision honest when the results are mixed.
What a European team can do now
Not being in the first wave is a reason to prepare, not to wait. The Android XR SDK and its glasses emulator are public, and on Android XR glasses an app can run on a host device such as a phone and project its interface to the glasses. That means the hard parts — the task flow, the data access, the permissions, the measurement — can be built and tested on phones your crews already carry, before a supported pair of glasses is available in your market. How the agent should be built across those layers is the subject of the companion piece, an agent on Android XR glasses is a four-part system.
A test you can run this week
Take the task somebody has proposed for AI glasses and fill in one line for each of the five items above. If you cannot name the current time and error rate for the task, measure them first; that measurement is the most valuable output of the whole exercise and costs nothing. If you cannot write a stopping condition everyone agrees on, the pilot is not ready, whatever the device can do.
Our part
In field deployment AI engineering we scope pilots like this one: on site, with one crew, with the data path, the measurement and the stopping condition written down before any hardware is ordered. We do not resell glasses or platforms, so the recommendation can be “not yet” as easily as “go”. More about how we work is on the home page.
- XR glasses
- on-device agents
- pilot
- field deployment