BeyondPlay is a social platform for making games with AI. You describe a game, an agent builds it, and minutes later your friends are playing it.

I build the whole thing myself. The iOS app, the web app, the backend, the design system, the agent that writes the games.

This is not a technical write-up. It is closer to a note about the decisions that shaped it, and about the small things I kept going back to fix.

Why social first

Smartphones and YouTube turned anyone with a phone into a media house. AI is doing the same for games, and most of the tools built on that idea stop at generation. You get a thing. Then you look at it alone.

Games have never worked that way. A game is only really a game once someone else is playing it with you, or after you, or against you. So the generation part was never the whole product to me. It is the on-ramp.

That is why the centre of BeyondPlay is a feed you play instead of watch. Full-screen games, one swipe apart, made by people you follow. Every game becomes something to beat, and every score becomes a reason to send it to someone.

From idea to playable

You tell the agent what you want in chat, and a ramble is completely fine. It works in its own cloud sandbox with a live preview running beside the conversation, so you watch the thing take shape instead of waiting on a progress bar.

The stream is durable. You can lock your phone in the middle of a build, come back an hour later, and the conversation picks up exactly where it was rather than starting over. That one took a long time to get right, and nobody will ever notice it, which is sort of the point.

When it is ready, publishing gives the game its own link. From there it can be played, remixed, put into a duel, or dropped into a group chat.

Details that took the longest

Most of my time did not go into features. It went into a handful of moments that decide whether the product feels alive or feels like a web page in a wrapper.

Streaming text fades in a word at a time as the agent thinks. It is animated on the UI thread, so it stays smooth even while the phone is busy building a game in the background. Text that appears in chunks feels like a machine reporting. Text that arrives like this feels like something thinking.

Games never assemble on screen. Before a game is revealed, it has to go quiet: frames settled, network settled, nothing still moving. Only then does it appear. You see a finished thing, never a half-built one.

The next game in the feed is already loaded before you get to it, without stealing the graphics thread from the game you are currently playing or leaking its audio into your ears. This is the single most fragile thing in the app and the reason swiping feels the way it does.

Duels are fair by default. When a game publishes, it gets seeded randomness, so two people in a head-to-head face exactly the same run of a game neither of them wrote. Fairness became a property of the platform rather than something each generated game has to remember to do.

Building it alone

There is no handoff here, which changes what design even means day to day. A decision about how a transition should feel does not become a spec, a ticket, and a negotiation. It becomes a commit.

The cost is that everything is mine, including the parts nobody thanks you for. The benefit is that the distance between noticing something is slightly wrong and it being fixed is usually an afternoon.

Where it is now

Fifty creators are in the closed pilot, a little over two thousand people are on the waitlist, and the iOS app is in App Store review.

Still early, and still the most fun I have had building anything.

If you want to try it before it is public, there is a TestFlight build. DM me on X and I will send you the link.

React Native (Expo) · Next.js · TypeScript · Vercel AI SDK · Cloudflare Workers