Roblox Studio Goes Agentic with Planning Mode, Mesh Generation, and Self-Testing AI Assistant
Roblox unveiled an agentic upgrade to its Studio Assistant on April 16, adding a Planning Mode, natural-language Mesh Generation, code-driven Procedural Models, and a playtesting agent that reads logs and uses inputs to find bugs.
Overview
Roblox is pushing its in-Studio AI Assistant from single-shot prompt completion toward end-to-end, agentic game development. On April 16, 2026, the company announced a new Planning Mode for Roblox Assistant, a Mesh Generation tool that drops fully textured 3D objects into a scene from natural language, and Procedural Models that let creators define editable 3D structures in code, according to TechCrunch. The update also adds a playtesting capability that uses simulated keyboard and mouse inputs and reads game logs to identify bugs and feed fixes back to the Assistant.
The announcement reframes Roblox Studio around a planning-first workflow rather than the prompt-and-pray pattern that has defined most generative AI tooling for game makers, and it lines up Roblox to interoperate with the same external coding agents that developers already run elsewhere.
What Roblox Announced
The centerpiece is Planning Mode, which Roblox describes as turning the Assistant into a collaborative partner that analyzes a game’s code and data model, asks clarifying questions, and translates prompts into editable action plans before any change is committed. Creators can review and tweak the plan, and the Assistant then executes against it and uses playtesting tools to verify the result, as TechCrunch reports.
Two new asset-creation features sit underneath Planning Mode:
- Mesh Generation produces fully textured 3D objects directly inside a game world from natural-language prompts, replacing placeholder assets during early development, according to TechCrunch.
- Procedural Models are editable 3D models defined in code and refined through the Assistant. Attributes such as the number of shelves on a bookcase or the height of a staircase can be adjusted dynamically and reused, creating parametric building blocks rather than one-off static meshes, TechCrunch reports.
Nick Tornow, Senior Vice President of Engineering at Roblox, framed the release as a broader shift in how the Assistant operates, moving from one-shot answers to multi-step workflows that plan, build, and check their own work, TechCrunch notes.
A Self-Testing Loop
The playtesting agent is what most clearly distinguishes the April 16 release from prior Studio AI features. According to TechCrunch, when Planning Mode executes against its plan, it uses playtesting tools to read output logs, capture screenshots, and feed back keyboard and mouse inputs to verify gameplay against the plan, then identifies bugs and lets the Assistant patch them automatically, TechCrunch reports.
The Next Web characterized this as a self-correcting loop in which the Assistant tests its own outputs, identifies problems, proposes fixes, and incorporates findings into subsequent planning cycles, as reported by The Next Web. For solo and small-team creators, that compresses a workflow that today typically requires manual QA passes between iterations.
Building on the Cube Foundation Model
The new Mesh Generation tool sits on top of Roblox’s existing 3D foundation model, Cube. In March 2025, Roblox open-sourced its 3D model Cube as the first launch of its GenAI tools, with Cube serving as the company’s foundational model for 3D generative AI creations and powering a Mesh Generation API inside Roblox Studio, according to Game Developer. The same Cube lineage was the basis for Roblox’s later 4D Generation feature, which lets players generate functional, interactive objects — such as a drivable car — from text prompts inside experiences.
In February, Anupam Singh, Roblox’s SVP of Engineering, described the Cube Foundation Model as adding “the dimension of interactivity” so that generated 3D objects “behave the way players expect,” according to Game Developer. The April 16 update extends that lineage by aiming the same generative pipeline at developers inside Studio, not just players in published experiences.
Early metrics from the player-facing version provide a baseline for adoption. According to The Next Web, more than 160,000 objects were generated during early access of 4D generation, and Roblox said players using the feature showed a 64% increase in play time on average, as reported by The Next Web.
Interoperability With External Coding Agents
A notable element of the announcement is Roblox’s intent to let creators use external coding agents inside Studio. The company said it wants creators to be able to seamlessly use Claude, Cursor, Codex, and other third-party tools with Roblox Studio, TechCrunch reports. The Next Web describes this as an MCP client integration enabling connections to those tools, alongside future plans for multi-agent parallel cloud workflows, according to The Next Web.
That positioning matters because it signals Roblox is choosing to ride the broader Model Context Protocol ecosystem — already adopted by Anthropic, OpenAI, and most major developer-tool vendors — rather than wall its developer audience into a single first-party assistant. It also puts the Studio Assistant in direct workflow competition with general-purpose coding agents that creators are already using.
Business Backdrop
Roblox is pushing the agentic update from a position of platform strength. Monthly active users grew from 280 million to 380 million through 2025, daily active users reached 144 million in Q4 2025, and full-year 2025 revenue was $4.9 billion, a 36% increase, with the company guiding to $6 to $6.2 billion in 2026, The Next Web reports. Total Robux purchases reached $6.79 billion in 2025, according to The Next Web.
Those numbers help explain why Roblox is investing heavily in lowering the barrier to creating new experiences: the platform’s growth depends on a steady supply of new games from solo and small-team creators who do not have dedicated engineering, art, or QA pipelines.
What We Don’t Know
Roblox has not published a firm general-availability date for the new Planning Mode and Procedural Models in Studio, the precise rollout schedule for the playtesting agent beyond a beta status, or pricing implications for creators relying on the new agentic workflows. It is also unclear how the Studio Assistant will mediate authentication and code execution when external tools like Claude, Cursor, and Codex connect via MCP — a question that has driven much of the security debate around agentic developer tooling across the industry over the past year. The April 16 announcement frames the technical pieces but leaves operational details for future updates.