AR welding has existed as a technology demo for years - twin high-framerate / high dynamic range cameras mounted on the forehead of the helmet, with periscope prisms to point forward, streaming back to VR optics in the helmet, provides the light adjustment of a traditional electronic welding glass, but also provides in-view guides for welding tool angle, arc length, travel speeds etc.
The real question is how useful this is, because it's primarily stuff that an experienced welder will already know, so the guides might just get in their way (in welding you're trying to reduce the amount of mediation between the welder's eyes and their workspace), and a welder who isn't experienced enough, probably shouldn't be working on anything that isn't a practice piece, so their learning might be better in a pure simulation.
That's the thing with AR - a lot of the things it's proposed as a solution for, aren't actually the problems the advocates think they are. This is pretty common when Great Technologist Saviours try to "solve" problems in ways that just so happen to require their contribution. The problem they're usually trying to solve is "how do I make (my thing) the solution to this problem?"
Look at education for a classic example - the only metric which has ever been demonstrated to improve educational outcomes for students, is fewer students per teacher. No laptop, tablet, electronic whiteboard, internet connection has ever produced a demonstrate improvement in educational outcomes.
Adding more teachers (or having fewer students) works consistently, but technology companies don't sell teachers, and laptops don't unionise and embarrass education ministers.
Crafts that also require a clear vision of the work surface, in which the knowledge of what needs doing is internalised. What the AR dream effectively comes down to, is people imaging an AR programme that can teach them a new skill while they perform the task. Unfortunately, there are very few tasks that anyone should be learning unsupervised on the real target.
For example, someone might think "safety glasses that have an AR overlay so your CNC mill can tell you how deep the hole you've drilled is, so you can get it right without taking your eyes off the workpiece", except you wouldn't do that live, you'd pre-configure the mill to stop the drill at the appropriate depth, etc.
Combat aircraft use them, and that's the real distinction. Combat aircraft use HUD / AR systems because they do things that no other aircraft have to do (save perhaps crop-dusting). There are actually very few use case scenarios where the need for information about the active work space is better deployed in a way that obscures the active workspace.
