— added by Claude · 2026: The original "post" was a single bookmark — a sub-page titled WebXR 入门必读 ("WebXR intro reading"), itself blank. A saved-for-later that never got written. Three years later the three technologies have settled into very different shapes, and the original framing — web platform vs. native app — has aged in ways worth writing about.
Three years on
— Claude in collab with Kai · 2026: Three web technologies, three different fates. Not the unified "web challenges native" wave 2022 was hoping for.
PWA — survived a near-death experience
The most dramatic story is the smallest technology. In February 2024, Apple shipped iOS 17.4 in the EU with Home Screen web apps disabled, citing DMA-driven security concerns about non-WebKit browser engines. The European Commission opened questions; developers and regulators pushed back; Apple reversed the decision within three weeks.
What actually got argued in those three weeks is what's interesting. PWAs are not a market — they are a capability. Which means they don't have a constituency that loses revenue when they disappear. They survived because a handful of vocal developers and regulators noticed in time. The 2022 framing was "PWAs will route around the App Store." The 2026 reality is more like: PWAs exist at the pleasure of the platform owner, who has now publicly tested how much it can take away.
WebXR — shipped, but only half of it
The Vision Pro launched in February 2024. visionOS 2 (September 2024) made WebXR available in Safari by default; Safari 26.2 (late 2025) added WebGPU integration so WebXR scenes can render with the modern GPU API.
But: only the immersive-vr mode. The immersive-ar module — the one that would actually let a webpage do passthrough AR on a device whose entire reason to exist is passthrough AR — is still not implemented in 2026. Meta's Quest browser supports it, with hand tracking, plane detection, anchors, and hit testing.
So the technology shipped, on the device that should have been its showcase, in a form that excludes the use case that device was built for. That is not the kind of arrangement you get by accident.
WebGL — replaced before it became dominant
WebGL never quite became the universal 3D layer of the web. While it was getting there, WebGPU shipped in Chrome (2023), Safari (2025), and Firefox stable (2025–26). As of January 2026 it counts as Baseline. Three.js and Babylon.js both have WebGPU renderers with WebGL fallback. Reported speedups land around 2–3× for graphics workloads and 15–30× for in-browser AI inference, where the new compute-shader path is doing the actual work.
The quiet thing about this: WebGPU is the only one of the three where the web platform's progress doesn't depend on a single vendor's mood. The standard came out of the major browser engines together, not as a concession from a gatekeeper.
What changed in the framing
In 2022 these three felt like a coordinated push — the web learning to install, render, and immerse like native. In 2026 they read more like three independent projects with three independent dependencies:
- PWA depends on the OS letting it exist.
- WebXR depends on the headset vendor caring enough to ship the modules.
- WebGPU depends on the GPU APIs underneath, which the browsers can target without asking permission.
The deeper shift: nobody really argues web vs. native as a binary anymore. The 5G performance gap is mostly closed for ordinary apps; agentic discovery (being indexable by AI agents) pushes toward web; retention and biometrics push toward native; most serious products do both. The 2022 question was "can the web replace the App Store?" The 2026 question is narrower and more honest: for which capability, on which platform, does the web get to be a first-class option this year? The answer is different for each of these three, and that's the whole story.