nobody talks about the thing actually killing xr adoption.
XR conversations love to jump straight to field of view, resolution, weight, and price. Those things matter. But I keep thinking about the person standing in a living room with a headset half-adjusted, repeating calibration for the third time and wondering why the future feels like troubleshooting.
Onboarding is where a device teaches people whether they are capable of using it. A good system notices confusion, explains the right thing at the right moment, and recovers without making the user feel at fault. Janu was my attempt to explore what that could look like with an AI assistant that responds to the actual setup problem instead of replaying the same generic instructions.
it’s not the price, it’s not the hardware. it’s the 10 minutes between opening the box and wanting to put it back in.
you can build the most advanced headset on the planet, but if setup feels confusing, people check out before the experience even begins.
i’ve watched smart people, including engineers, designers, friends, and family, struggle with headset setup. those repeated calibration failures. improper ipd causing blurred vision. gestures that don’t respond. instructions that repeat themselves without understanding the problem.
eventually, they ask me, “why are you still doing xr?”
that’s the moment the device loses a user.
i’ve been thinking about this problem for six months, so i built something.
meet janu🐘, it's an open-source research prototype exploring what ai-assisted onboarding for xr could look like.
the core idea is simple: > what if xr onboarding didn’t repeat the same instructions? > what if it could recognize when someone is struggling, adapt its guidance, and show developers where the experience is breaking
during onboarding, janu observes interaction signals such as repeated commands, timeouts, and reported failures to estimate when someone is struggling.
for developers, it turns those sessions into actionable insights: > where users struggle or stop progressing > why janu intervened at a specific step > which steps fail across multiple sessions > what to fix before more users face the same problem
so i built the full stack with an agent, confidence scoring, session replay, analytics dashboards, and a Unity demo.
What I remember most is the shift from an idea in my head to something another person could actually touch. Shipping makes the questions sharper, the feedback more honest, and the next version much easier to imagine.