Epic put together Unreal Engine 5.9 Preview with AI agents in about a week

A coding agent can write a script for Unreal Engine, but developers often still have to bring the result into the editor themselves, run it and troubleshoot errors. In Unreal Engine 5.9 MegaJam Preview, Epic is expanding the tools that let an agent act within a project. One practical use case is to ask it to assemble a rough scene and check the result before committing an artist's time to the details.

The build is available for experimentation. Its main draw is the combination of agent access to the editor, a generative canvas and browser deployment. For a team, this offers a way to test several stages of prototype production at once, from modifying a scene to presenting it to a client.

The agent gets editor tools

The Unreal Assistant panel lets developers connect their usual coding agent. Unreal MCP acts as the connecting layer, giving an external program access to permitted editor actions. The documentation lists tools for Blueprints, files, scene capture, profiling and other tasks. Expanded Python and Blueprint libraries allow graphs and variables to be modified programmatically. The available toolsets are managed through Toolset Registry. Epic's overview of the changes.

Imagine a team building a virtual showroom demo. A developer asks the agent to add an interactive display stand, set up the logic for opening a product card and capture an image of the scene. This is an example of a possible assignment, not a published test: execution quality still needs to be measured on a specific project.

This kind of integration should be assessed by the time it takes to reach a verified result. How many revisions were needed after the first run? Does the project build? Can another developer understand the Blueprint it created? If an agent quickly produces a structure that then takes a long time to fix manually, the team is simply shifting work between stages.

Canvas connects generation to the working scene

Unreal Canvas brings text, images, video and mesh generation together on a canvas. The Canvas Bridge plugin provides the canvas with rendering through Movie Render Graph, while Spatial Annotations lets users place annotations in a scene and capture it for further work. This provides a basis for discussing a specific shot alongside different ways of processing it. Release Notes.

For the showroom, this workflow could be tested on a single display: prepare the source shot, generate several visual variations and choose a direction. Then assess separately which changes are suitable for an actual interactive scene. A successful image alone says nothing about the suitability of the geometry, materials or rendering speed.

Getting connected requires setup. According to Epic's instructions, local processing stages require an NVIDIA graphics card with at least 8 GB of VRAM. For external generation, users supply their own API keys, and the selected provider bills them. Installation also includes the Canvas server and a download of around 1.8 GB of local models. Getting started guide.

A week of development calls for careful testing

The claim that “the engine was written in a week” needs qualification. According to Epic, a small group of engineers supervised agents that ported AI tools from the UE6 branch and implemented the web platform, using UE4's old HTML5 platform as a starting point. Engineers directed the work, reviewed the code and tested the results. This does not mean that the entire Unreal Engine was rebuilt from scratch. Epic's announcement.

Browser deployment uses WebAssembly and WebGPU, but currently relies on the mobile renderer. The announcement specifies a source build, a C++ project and a preference for Chrome. Epic also warns that the Preview underwent reduced testing and that parts of the web renderer are prototypes. Promising that any game can simply be brought to the browser would therefore be premature. Preview requirements and limitations.

Developers should set aside a copy of a small project and one repeatable task. Product managers should define a success criterion in advance: for example, how long it takes to prepare a demo, including fixes and generation costs. That measurement will show whether the integration helps teams deliver verifiable prototypes faster.