AI agents, a creative workspace, and web export in Unreal Engine 5.9 Preview

Epic Games prepared this experimental release for its MegaJam. AI agents helped develop the preview and wrote the code for its new web target.

Unreal Engine’s return to the browser starts with an old piece of the engine. To develop the new web target, Epic Games used code from the retired Unreal Engine 4 HTML5 platform as a starting point. According to the MegaJam Preview announcement, AI agents wrote the code for the target under the direction of engineers, who reviewed and tested their work. A small team also directed agents as they ported AI tools from the Unreal Engine 6 development branch to version 5.9.

Epic describes this as the first Unreal Engine release prepared primarily with AI agents. The team put the preview together in about a week, starting from a branch where rendering and animation changes for version 5.9 were already underway. The quality assurance period was shorter than usual. Epic says it is still revising some of the code written by agents and plans to rework parts of the web renderer before the final release. Running an Unreal project in the browser

The experimental target combines WebAssembly to run compiled code in the browser and WebGPU to access the device’s graphics hardware. Once a project has been built for the web, players can open it directly in a compatible browser without installing Unreal Engine. For developers, the current preview requires a C++ project and a build of the engine compiled from source.

Epic recommends Chrome for this preview. Web builds use Unreal’s mobile renderer, which determines which graphics features are available. Epic also says cooking, the step that prepares project assets for the target platform, can take longer and that performance may vary. What an agent can do inside the editor

Already documented in version 5.8, Unreal MCP exposes editor commands that an agent can call. Depending on which tools are enabled, an agent can inspect selected objects in a scene, add a light, edit a Blueprint, or run a test. The preview expands that access to areas including performance profiling, geometry, and audio. The Fab integration lets agents search the catalog, browse the account’s library, and import assets into the project. The expanded Python APIs also make more operations available to scripts, including creating and editing Blueprint graphs.

The Unreal Assistant panel is where developers chat with an agent. It can launch Claude Code, Codex, Cursor, or OpenCode and connect them to the editor’s MCP server. The external agent uses the developer’s account with its provider. What it can do depends on which tools are enabled in Unreal. The panel is where the developer makes the request, and MCP gives the agent a way to act on it inside the project. Unreal Canvas starts with the 3D scene

With Unreal Canvas, a scene already set up in the editor can serve as a reference for generation. A team can start with a camera framing, send a capture, and, depending on which tools are enabled, include depth data, object masks, or annotations placed in the scene. When exploring several looks for the same shot, those inputs give the generative model a reference for the composition and placement of objects.

The workspace keeps references, prompts, and results together. Teams can generate images, videos, or 3D meshes, then import supported assets back into the project. The setup guide explains that teams connect their chosen services using their own API keys and that each provider bills for generation. In the documented setup, local scene analysis tasks such as segmentation and depth estimation require an NVIDIA GPU with at least 8 GB of video memory.

The preview is available through the Epic Games Launcher, as a Linux download, and through the Unreal Engine GitHub repository for authorized accounts. It arrives ahead of the Epic AI MegaJam, November 12–15, where participants can build a game in Unreal Engine or Unreal Editor for Fortnite.