Vibe-Coded Android Apps Now Run Privately Through Google Play

From prompt to phone, without the developer toll booth

Google has made it much easier to get an AI-generated Android app onto a phone. On October 13, the company announced that anyone with a standard Google account can describe an app in Google AI Studio, have it built, and then launch it on their own device through Google Play. A paid developer account is not required.13 These personal apps sit in a private part of Google Play and are left out of the public catalog.11 The rollout is starting in the United States and some other markets, and only users 18 and older can use it.23

The idea is simple even if the engineering behind it is not. A user writes a plain-language prompt, and an AI agent produces a native Android project. While the app is being built, the tool suggests features the user may have forgotten to ask for. Afterward, the user can make small edits or throw the app away and start over.11 Building and experimenting cost nothing.11 Users skip the $25 Play Console registration fee that has long been the entry cost for Android distribution.7

The reports agree on the main facts. They differ on how big a change this is, and that difference is worth explaining.

The building came first; this release fixes delivery

Most headlines treat prompt-based Android creation as new. It isn't. The Verge notes that Google first let people vibe-code personal Android apps in May, and what changed now is that users no longer have to plug their phone into a computer to install them.10 Other accounts tie the Android option in AI Studio's Build mode to Google I/O in late May 2026. They disagree slightly on the date, with one saying 19 May and another saying 20 May.184 Google's Logan Kilpatrick has reportedly said users built more than 250,000 Android apps in the first week.18

The obstacle was installation. Until this update, putting a generated app on a phone meant using ADB, a developer tool that most non-programmers have never used. The alternative was publishing to Google Play, which would expose the app to the store's roughly three billion users.5 AI Studio did have an app creator before, but it was meant for professional developer accounts and required USB debugging from a PC.11 The new path turns that into a one-click install on a personal device.11

The key point is that this is a distribution change, not a jump in what the model can build. AI coding tools have been able to generate working apps for some time. The hard part for ordinary users has been everything after the code is written: signing, sideloading, developer settings and fees. Google now controls all of those steps in one flow, and that is the actual product here.

What the agent actually produces

Google's developer documentation describes the technical pipeline. A prompt in Build mode causes the Antigravity Agent to generate a full Kotlin and Jetpack Compose project. The app can be previewed in an Android emulator that runs in the browser, installed on a physical device, or published for testing.12 The documentation also says AI Studio's Build mode can produce full-stack web apps, and that Android is a separate option in a platform picker.14 Users can revise apps through chat or through an annotation mode, where they highlight part of the interface and describe the change they want.14

Since the output is native Kotlin rather than a web wrapper, it uses the same tools professional Android developers use. That gives Google an edge over many rival builders, which typically generate React Native or other cross-platform code.1617 One hands-on test built the same habit tracker several ways. The Android version took 14 minutes and 17 seconds from a single prompt and produced about 25 Kotlin files.4

Where the coverage diverges: capabilities and limits

The reports disagree most about what these personal apps are allowed to do. ExtremeTech says the apps can use the camera, GPS, accelerometer and Bluetooth, and can call Gemini APIs. It also lists firm limits: only client-side features, no server-side services such as Firebase, single-activity and single-module projects only, and no Wear OS or Android TV.7 Appy Pie's testing found the same thing, reporting that Android builds had no database, sign-in or secrets because they run entirely on the phone.4

Google's own Android documentation, updated October 20, says something different. It describes Firebase integration that can call Gemini through Firebase AI Logic and App Check, store data in Cloud Firestore and sign users in with Firebase Authentication. It also says the agent can provision those services automatically when a prompt asks for them.12 The most likely explanation is that the feature set is changing quickly and some of the hands-on reviews were done on an earlier version. Readers should treat the official documentation as the most current source and expect some features to behave differently depending on account type or market.

The reports also differ on how open distribution is. Engadget describes the change as letting people distribute vibe-coded apps through the Play Store with only a Google account.2 In practice, at launch the apps were available only on the creator's own devices, through a new personal library section.2 Sharing with friends and family is promised for the coming months.105 Reaching a wider audience still means moving up to a professional developer account and going through the Play Store's normal safety reviews.2 Even the existing internal-testing route, which is capped at 100 testers, needs the $25 developer account, and one review found AI Studio's publish panel labeling public Play Store publishing as "coming soon."4 For now, "distribute" is a strong word for something that mostly lets you install your own app.

The quality question vibe coding hasn't answered

Making app creation easier doesn't make the apps any better. Engadget notes that how useful these features are depends on how much users trust AI to build software.2 The most detailed independent test supports that caution. AI Studio's web version of a habit tracker had bugs the testers had to find on their own, and one of the agent's fixes introduced a new bug.4 Even after a round of fixes, bad data that had already been saved stayed in place and an impossible date was still accepted.4 The same reviewers noted that AI Studio did not say what it had not tested, while Claude Code did.4

There are practical risks for people who share what they build. Google's documentation says a deployed web app uses the creator's API key for every user's Gemini calls, so other people's usage counts against the creator's limits and costs. People you share an app with can also see its code and fork it.8 Free-tier users should also be aware that Google may use what they type to improve its products.4 All apps, private or not, still have to follow Google Play's existing policies.1

These issues are why the private-first design makes sense. A personal app with a few rough edges harms almost no one. The same app shared with relatives or published to the store creates a support and security burden that a non-developer may not see coming. Google seems to understand this and has kept wider sharing behind a security review and the standard developer process.37

Why Google is doing this now

There are two reasons, and they support each other. The first is competition among AI coding platforms. Many startups now promise native mobile apps from a prompt, including CatDoes, Rork, Adalo, Thunkable and Emergent.161915 Most of them charge to publish to app stores. CatDoes, for example, lists store publishing on a $50-per-month plan, and Thunkable starts at $37 a month billed yearly.13 Google offers free generation of native Kotlin apps plus one-click installation, which removes much of the reason to pay a third party for personal-use tools. One analysis says independent app-builder startups will now need a sharper reason to charge for a similar result.3

The second reason is the condition of the Play Store itself. Engadget places the move in the context of legal pressure to open the store to competitors and argues that it shows Google trying to stay competitive with third-party app stores.2 Google also made sideloading harder over the past year.2 Seen that way, a sanctioned and policy-checked route for homemade apps gives Google an answer to critics: people can run their own software, as long as it goes through Google.

The takeaway

This is a true step toward AI-first software development, but the change is in logistics rather than intelligence. Building an Android app from a prompt was already possible. Installing one without ADB, a developer account or a fee is what's new. That opens the door to a group of people who want narrow tools, such as ad-free utilities, single-purpose trackers or app ideas too specific for any store, and who never wanted to become developers.117

What comes next will depend on the features Google has not shipped yet. Wider sharing, more markets and public publishing are where the real tests of quality, security and moderation will happen. Until then, the advice from people who have tested these tools still applies: use the app the way a real user would, and test it again after every fix the AI makes.4