Android Instant Apps / The Slow, Inexorable Death of HTML

Android Instant Apps were announced at the recent Google I/O. Based upon available information1, Instant Apps offer the ability for websites to instead transparently open as a specific activity/context in an Android App, the device downloading the relevant app modules (e.g. the specific fragments and activities necessary for a need) on demand, modularized for only what the context needs.

The querystring app execution functionality already exists in Android. If you have the IMDB app, for instance, and open an IMDB URL, you will find yourself in the native app, often without prompting: from the Google Search app it is automatic, although on third party sites it will prompt whether you want to use the app or not, offering to always use the association.

Click on that link in newer versions of Android (in a rendering agent that leverages the standard launch intents), with IMDB installed, and you’ll be brought to the relevant page in that app.

Instant Apps presumably entail a couple of basic changes-

  • Instead of devices individually having a list of app links (e.g. “I have apps installed that registered for the IMDB, Food Network and Buzzfeed domains, so keep an eye out for ACTION_VIEW intents for any of the respective domains“), there will be a Google-managed master list that will be consulted and likely downloaded/cached regularly. These link matches may be refined to a URL subset (where the current functionality is for a full domain).
  • An update to Android Studio / the build platform will introduce more granular artifact analysis/dependency slicing. Already this exists in that an APK is a ZIP of the various binary dependencies (e.g. for each target processor if you’re using the NDK), resources, and so on, however presumably the activities, classes and compiled resources will be bifurcated, their dependencies documented.
  • When you open a link covered by the master list, the device will check for the relevant app installed. If it isn’t found, it will download the necessary dependencies, cache them in some space-capped instant app area, initialize a staged environment area, and then launch the app.

They promise support, via Google Play Services, all the ways back to Android 4.1 (Jellybean), which encompasses 95.7% of active users. Of course individual apps and their activities may use functionality leveraging newer SDKs, and may mandate it as a minimum, so this doesn’t mean that all instant apps will work on all 95.7% of devices.



The examples given include opening links from a messaging conversation, and from the Google Search app (which is a native implementation, having little to do with HTML).

The system will certainly provide a configuration point allowing a device to opt out of this behavior, but it clearly will become the norm. Google has detailed some enhanced  restrictions on the sandbox of such an instant app — no device identification or services, for instance — but otherwise it utilizes the on-demand permission model and all of the existing APIs like a normal app (detailed here. As is always the case, those who don’t understand this are fear mongering about it being a security nightmare, just as when auto app-updates were rolled out there were a number of can you say bricked? responses).

And to clear up a common misconception, these apps are not run “in the cloud”, with some articles implying that they’re VNC sessions or the like. Aside from some download reductions for the “instant” scenario (Instant Apps are apparently capped at 4MB for a given set of functionality, and it’s tough to understand how the rest of the B&H app fills it out to 37MB), the real change is that you’re no longer asked — the app is essentially forced on you by default — and it doesn’t occupy an icon on your home screen or app drawer. It also can’t launch background services, which is a bonus.

Unfortunately, the examples given demonstrate little benefit over the shared-platform HTML web — the BuzzFeed example is a vertical list of videos, while the B&H example’s single native benefit was Android Pay — though there are many scenarios where the native platform can admittedly provide an improved, more integrated and richer experience.

It further cements the HTML web as a second class citizen (these are all web service powered, so simply saying “the web” seems dubious). I would cynically suggest that the primary motivation for this move is the increased adoption of ad blockers on the mobile HTML web: It’s a much more difficult proposition to block ads within native apps, while adding uBlock to the Firefox mobile browser is trivial, and is increasingly becoming necessary due to the abusive, race-to-the-bottom behaviors becoming prevalent.

And it will be approximately one day before activities that recognize they’re running as instant apps start endlessly begging users to install the full app.

Ultimately I don’t think this is some big strategic shift, and such analyses are usually nonsensical. But it’s to be seen what the impact will be. Already many sites treat their mobile HTML visitors abusively: one of the advocacy articles heralding this move argued that it’s great because look at how terrible the Yelp website has become, which is a bit of a vicious cycle. If Yelp can soon lean on a situation where a significant percentage of users will automatically find themselves in the app, their motivations for presenting a decent web property decline even further.

1 – I have no inside knowledge of this release, and of course I might be wrong in some of the details. But I’m not wrong. Based upon how the platform is implemented, and the functionality demonstrated, I’m quite confident my guesses are correct.