Essays
Longer pieces about where AI and XR actually meet, and where the seam between them is doing more work than the pitch admits.
The runtime gives you geometry, not meaning
Scene understanding is the obvious use for a vision model on a headset, and it is also where the web platform has drawn a line it does not intend to move.
9 min readThere is no typing indicator in a room
Conversational latency that a text interface absorbs invisibly becomes a performance problem the moment the speaker has a body.
10 min readWhat sits between a generated mesh and a WebXR scene
The generation step got fast. The steps on either side of it did not, and the gap between them is where the time actually goes.
10 min read
Why these are separate from the topic pages
A topic page is pinned to an interface. It answers what a call does, what the runtime does with it, and what goes wrong when you get a detail subtly off. That format works because the question has an answer you can check against a specification.
These pieces are about questions that specification cannot settle, which mostly arise when two technologies are pushed together and someone has to decide which one gives way. Whether a generated mesh belongs in a frame budget, what a two-second silence means when the speaker has a body, how much a runtime is willing to tell a web page about the room it is in — none of those are API questions, and answering them in an API reference would be the wrong shape.
They are opinionated in a way the topic pages are not, and they say so where a claim is a judgement rather than a measurement. Where a number appears it comes with its source, because the fastest way to make an argument like this useless is to invent a benchmark for it.