Building The Next Era of Creation on Roblox

Why Games Are Unique in the Age of AI

A game is a system, not a file. A game is hundreds of things that depend on one other: scripts, physics, UI, server logic, and multiplayer state. Change how high a character jumps, and suddenly a level can be skipped, a puzzle stops working, and the server has to sync something new. AI can generate a first draft quickly, but getting all those parts to work together takes a loop: change, run, observe, fix, repeat.

Games are verified by play. You can’t tell whether a game is fun by reading its code or looking at a screenshot. You have to play it—then thousands of other people have to play it. Unlike text, images, or video, a game can’t be judged as output; it has to be experienced. That’s why AI game creation needs a real runtime, real playtesting, and a real player signal to get it right.

Games are social and live. The best games aren’t finished at launch. They’re operated: updated weekly, tuned to player behavior, and sustained by communities of friends who play together. Generation is the beginning of a game’s life cycle, not the end.

Abundance shifts the bottleneck. When anyone can make a game in an afternoon, making one stops being the hard part. The hard part becomes getting it in front of the right players, earning their time, and giving them a reason to come back with friends. As creation gets cheaper, distribution, discovery, and community become scarce resources that can’t be generated.

Roblox’s Unique Ecosystem

One integrated stack. Roblox’s engine, Studio; cloud services; game servers; distribution; economy; and safety all live on the same platform. That’s what makes Roblox such a strong environment for agents: they can act on and observe the entire life cycle of a game, not just one slice of it.

A built-in audience on day one. Every day, 138 million users arrive with their friends, identities, and communities already in place. AI can generate a game in an afternoon. It can’t generate an audience of that size or the infrastructure that serves it.

Discovery that rewards quality. Our discovery system is designed to surface games based on how players engage with them over time, so that great games rise and low-engagement content falls, no matter how much new content arrives or how that content is made.

A real business for creators. Built-in commerce, our Marketplace, and the Developer Exchange (DevEx) program let creators go beyond publishing to earning from what they build.¹

Trust at scale. Safety and moderation tools, combining technology and human review, are built into the platform and help protect players and creators alike. 

The faster creation gets, the more these advantages compound.

Under the Hood: The Harness, MCP Server, and Build

We’re building AI into every layer of creation on Roblox, for every kind of creator.

The game-building harness. An agent is only as good as the environment it can act in. Our harness gives agents a real game runtime to work against and is model agnostic, so we can adopt the latest internal or frontier models as soon as they’re the best tool for the job. Unlike most vibe-coding tools, it runs on the same process we use to launch a Roblox game server, which makes it faster and more resource-efficient. The agent isn’t simulating a game. It’s running one.

A continuous creation loop. In Roblox Studio, creators can connect an agent, make changes, run a playtest, inspect the logs, and repeat until the game works the way they intended. Then they can deploy updates to every device and region in seconds. That loop of generate, play, observe, and fix is exactly what game creation requires, and it’s native to Roblox.

The Studio MCP server. Studio power users can bring the AI agent of their choice, such as Claude Code or Codex, and securely connect it to their Studio session through our server built on the open Model Context Protocol (MCP). Since June 30, time spent using AI tools has tripled among our top creators, and daily active users of the MCP server have grown nearly fivefold.