Meta’s Pocket turns AI app building into a social media experience

Meta is testing a new kind of social network where the content users scroll through is not just something to watch or read. It is something they can play with.

The company has rolled out Pocket, a standalone app that lets people use natural-language prompts to create small interactive experiences called “gizmos,” without needing to write the code themselves.

Users can then play with those creations, share them through a scrollable social feed and remix experiences created by other people.

That makes Pocket more than another AI-powered creation tool. It is an experiment in turning vibe coding into a social activity.

From prompting AI to creating something people can play

Pocket allows users to describe what they want to create in ordinary language, with the system generating the interactive experience.

The resulting gizmos are designed as lightweight, mobile-first experiences rather than full applications or commercial software products.

They can respond to actions such as tapping, dragging and swiping, as well as the phone’s movement through its accelerometer. Some experiences can also use sound and access capabilities such as the camera, microphone or photo gallery.

That makes the experience different from conventional AI assistants, where the primary output is usually text, an image or another piece of media.

With Pocket, the output is something the user can interact with.

Meta’s own Help Center describes a gizmo as an “interactive, playable AI-generated experience.”

The distinction is important.

The company is not trying to turn every Pocket user into a software engineer. It is reducing software creation to the same kind of casual interaction that social platforms have already made familiar with photos, videos and memes.

The social feed may be the bigger idea

The most interesting part of Pocket may not be its ability to generate a mini-game.

It is what happens after the game is created.

Pocket includes a scrollable feed where users can discover gizmos made by other people. They can play them, save them, share them and create their own versions.

Users can also remix existing gizmos rather than starting from a blank prompt.

That creates a loop that looks more like social media than traditional software development:

Create → play → share → remix → create again.

This is a meaningful shift from how AI coding tools have traditionally been positioned.

Most AI coding products are designed around productivity. A user has an idea, describes it to an AI system and uses the resulting code to build a website, tool or application.

Pocket makes the creation itself part of the entertainment.

A user does not necessarily need to have a business idea or a software problem to solve. They can simply make something interesting and put it into a feed for other people to experience.

In that sense, Meta is testing whether software can become another form of social content.

Meta is not building another app store

Pocket’s gizmos should not be confused with full applications.

The research indicates that the platform is designed for lightweight, prompt-native experiences rather than SaaS products or conventional commercial apps.

That limitation is also part of the product’s appeal.

Creating a full application can involve product design, coding, testing, deployment and distribution.

Pocket compresses that process into something much closer to creating a social post.

The result does not need to become a serious software product. It can simply be a small game, interactive joke, toy or experiment that someone else can discover in the feed.

That lowers the threshold for participating in software creation.

The technology behind Pocket

Pocket is part of Meta’s broader push into AI-assisted software development, with the company’s Muse Spark model family providing the underlying capabilities for coding and multimodal reasoning.

Meta Superintelligence Labs developed Muse Spark, with successive versions released during 2026.

The technology is important, but it may not be the most significant part of Pocket’s strategy.

Meta already has increasingly capable AI models.

What Pocket does is put those capabilities into a consumer product built around a new behaviour: prompting software into existence and immediately distributing it socially.

That could give Meta another way to understand what consumers actually want to create with generative AI.

Pocket grew out of Gizmo

The app also has a direct connection to an earlier startup.

Pocket is heavily based on the work of Atma Sciences, the company behind an earlier product called Gizmo.

Atma Sciences has said that it is winding down and that its team has moved to Meta.

This is important because some reporting has described the development as Meta acquiring Gizmo, but the available evidence points instead to a team transition and technology integration rather than a conventional acquisition of the company as an ongoing business.

The original Gizmo product reportedly accumulated more than 635,000 lifetime installs before the team moved to Meta.

Pocket therefore appears to be Meta’s attempt to take that underlying concept and build it into a larger consumer experiment.

Meta is moving beyond the AI chatbot

Pocket also fits into a broader change in Meta’s consumer AI strategy.

Meta continues to put its AI assistant inside major products such as WhatsApp and Instagram, but it is also experimenting with standalone AI experiences.

Pocket and Vibes point toward a strategy in which Meta does not simply add AI to existing social products.

It creates entirely new products around what AI makes possible.

That distinction could become increasingly important as AI models become capable of generating more than text and images.

If an AI can generate an interactive experience almost instantly, the next question is not simply what the user can generate.

It is where those creations go and what other people do with them.

Meta already understands the distribution side of that equation better than most technology companies.

Pocket gives it a way to test whether the social-media model can be applied to AI-generated software.

The next creator economy may be interactive

For creators, the significance of Pocket is potentially bigger than the app itself.

The internet has already moved through several stages of digital creation.

People first created websites and blogs. Then social platforms made photos and videos easier to publish. Short-form video eventually turned ordinary users into creators of entertainment distributed to millions of people.

Generative AI has lowered the cost of producing those forms of media even further.

Pocket points toward another possibility: interactive content becomes just as easy to create as video.

A creator could eventually make a small game, interactive meme or digital toy with a prompt, publish it immediately and let other users remix it.

That does not mean Pocket has already created that future.

The product is still in a regional rollout, and its current experiences are deliberately lightweight.

But the direction is worth watching.

The important change is not that AI can now write code.

Developers have been using AI coding tools for some time.

The more significant change is that Meta is experimenting with making the creation, distribution and consumption of AI-generated software happen inside the same social environment.

That turns software from something people download after it has been built into something that can become social content in its own right.

And if that model works, the next generation of digital creators may not only ask AI to generate something worth watching.

They may ask it to generate something worth playing.