Roblox’s AI assistant gets new agentic tools to plan, build, and test games

Roblox is fundamentally upgrading its Roblox Assistant from a simple prompt-to-code tool into an agentic, collaborative development partner. The centerpiece is a new 'Planning Mode' that addresses the common failure of one-step AI tools to capture creator intent. Instead of executing immediately, the Assistant now analyzes a game's code and data model, asks clarifying questions (like choosing a visual style or asset source), and translates prompts into detailed, editable action plans. Creators can refine these plans extensively before any code is written or assets are built, ensuring the final output matches their vision.

Once a plan is finalized, Planning Mode leverages Roblox's suite of AI building tools, including two new ones announced alongside it: Mesh Generation and Procedural Model Generation. Mesh Generation allows creators to instantly generate fully textured 3D objects (meshes) to replace low-quality placeholders. Procedural Model Generation enables the creation of editable, code-based 3D models where attributes (like shelf count or staircase height) can be dynamically adjusted. Crucially, the Assistant's role extends into testing; it can simulate playtesting by reading logs, capturing screenshots, using virtual inputs, identifying bugs, and providing feedback to automatically fix issues, creating a self-correcting development loop.

Looking ahead, Roblox is working on enabling multiple AI agents to collaborate on parallel tasks and run complex cloud workflows. The company also aims to ensure seamless integration with third-party tools like Claude, Cursor, and Codex within Roblox Studio, further breaking down barriers between creative vision and technical execution.