Roblox’s AI assistant gets new agentic tools to plan, build, and test games | TechCrunch
Short summary
Roblox rolled out a Planning Mode for its AI Assistant that breaks down development into editable action plans, asks clarifying questions, and integrates directly with code and data models. It also unveiled Mesh Generation and Procedural Model Generation to speed up asset creation, plus automated playtesting to catch and fix bugs. These updates aim to turn prompts into multi-step workflows for planning, building, and testing games.
- Roblox's new Planning Mode turns a single prompt into an editable, multi-step action plan by reading existing code/data and asking clarifying questions before building.
- Mesh Generation and Procedural Model Generation let creators produce fully textured, scriptable 3D assets (e.g., adjustable bookcases or stairs) instead of manual placeholders.
- Built-in automated playtesting logs output, takes screenshots, simulates inputs, and lets Assistant self-correct by looping fixes back into the plan.
- Roblox is adding support for parallel cloud-based AI agents and plans to integrate third-party tools like Claude, Codex, and Cursor directly into Studio.
Full summary
Roblox just overhauled its Assistant tool in Studio, turning it into a multi-step collaborator rather than a one-off generator. The new Planning Mode reads your existing code and data model, asks follow-up questions, then drafts an editable action plan. You can tell it to “build a park mini-game with a fountain and foliage,” and it’ll ask whether you want a cartoony or realistic style, or if you’d rather use Creator Store assets versus scratch builds. Once you lock in those choices, Planning Mode hands off tasks to specialized AI features.
Two of those features are Mesh Generation and Procedural Model Generation. Mesh Generation spits out fully textured 3D objects on demand—think a detailed campfire that already glows at night—so you don’t have to slap in blocky placeholders. Procedural Models go deeper, letting you script and tweak objects in code: change the number of bookcase shelves or the height of stairs with simple prompts, then reuse those adjustable assets anywhere in your game. Assistant tracks relationships in 3D space, so if you say “move the torch two meters above the doorway,” it knows exactly where and how to place it.
Testing is built in. As Assistant carries out the plan, it uses playtesting tools to log output, snap screenshots, even simulate keyboard and mouse inputs. It flags bugs, suggests fixes, and can loop back to adjust the plan on the fly. That self-correcting loop means each iteration learns from the last one, improving accuracy over time without a human rewriting every step.
Behind the scenes, Roblox is also wiring in support for multiple AI agents running parallel workflows in the cloud. Long coding or testing tasks can be farmed out, and you’ll eventually hook up third-party engines—Claude, Codex, Cursor—directly inside Studio. The goal is clear: shrink the gap between what creators imagine and what ends up in players’ hands.