Artificial intelligence has quietly become a core part of how the biggest games are made on Roblox. According to the company, 44% of the platform’s top 1,000 creators already use Roblox Assistant or third-party AI tools through MCP to plan, build and test their games. That’s a pretty surprising number for a feature set that has so far mostly lived on the fringes of workflow development.
Now, Roblox wants to bring artificial intelligence to the surface. The company is introducing a new wave of agent features in Roblox Studio and Assistant, as well as a built-in MCP server that opens up Studio to third-party AI tools. The goal is simple: creators should be able to use AI at every stage of the plan, to connect, build and test it in one step.
A smarter Planning mode that actually asks questions
Anyone who has spent time with existing AI coding tools knows the pattern. You write a query, the tool pops up a solution, and half the time what comes back looks like what you actually want. For something as layered as game development—where you’re juggling code, assets, systems, and design intent all at once—this one-shot approach quickly falls apart.
Roblox’s answer is an improved Scheduling mode that turns the Assistant into something closer to collaboration than an automaton. Before you touch anything, Assistant analyzes your game’s code and data model, asks clarifying questions, and works with you to break down a complex request into a detailed action plan that you can actually review and edit. Think of it less like a line of command and more like a back-and-forth with a small developer who wants to make sure they’re building the right thing before launching.
This plan then acts as a kind of mini design document. Agents can refer to it to execute tasks in parallel and check their work against the original vision, and Assistant maintains a structured manifest of tasks and contexts it can take across sessions. Roblox says the cross-session memory feature will come shortly after launch, making it more useful for larger, long-term projects.
New Mesh and Procedural Model tools to really build the world
When a plan is closed, the next bottleneck is asset creation – and that’s what Roblox is aiming for. Mesh Generation allows creators to release textured meshes into the game world quickly, reducing one of the most tedious parts of world building.
What’s more interesting is the following: Code-driven procedural models. Creators will be able to create these models from text or image prompts, or create them from scratch and then edit their attributes on the fly. The number of shelves in a bookcase, the number of chairs around a table, how the structure scales between different rooms – all of this becomes editable and reusable. This is a significant shift from static assets to intelligent, modular building blocks that can be refined and redefined as the project evolves.
A Playtesting Agent that handles QA for you
Testing is a part of development that most creators would happily download if they could, and Roblox’s new Playtesting Agent beta aims to do exactly that. The agent tests your game against the original plan, examines the code and data model, reads logs, and even uses the player character as an automated QA tester to verify that things are behaving as intended.
The bigger vision here is clear in a quote from Malt, creator of Solo Hunters: In the future, community members can spot bugs or feature requests, and my AI system can review and complete tasks overnight. When I wake up, all I have to do is check the pull requests and see what I want to integrate into the game.”
This is a far cry from the copy-and-paste AI workflows of even a year ago. Given that nearly half of Roblox’s biggest creators rely on AI tools for faster shipping, it’s not hard to see where this is going. The platform isn’t just adding AI features, it’s currently reshaping Studio.
