QORPO Blog / Industry

Unreal Engine 5.9 MegaJam Preview: what to test with an editor agent

Q
QORPO Studio17 November 2026 · 4 min read
A dark workstation with a small illuminated door model
Concept illustration created for this article, not a gameplay screenshot.

Epic announced the Unreal Engine 5.9 MegaJam Preview on November 9, 2026. It includes an experimental in-editor assistant and expanded Unreal MCP tools. A useful first evaluation is to copy a small project, have an agent make one reversible change and compare the resulting asset diff with what happens in a playable build.

What did Epic announce?

The official preview announcement ties the build to the November 12 to 15 Epic AI MegaJam. Epic says agents contributed to porting AI tooling and implementing the experimental Web target, with engineers directing, reviewing and testing that work. It describes the preview as having a shorter QA pass than usual and recommends testing on a copy of your project.

The in-editor assistant is one part of the build. Epic also describes experimental web support using WebAssembly and WebGPU, which currently requires a source build and a C++ project. That is a separate workflow with its own build and browser conditions. If you only want to understand how an editor agent changes a level, begin with the editor and a small local project. The QORPO MegaJam preparation guide puts the preview in the context of delivering a playable entry.

Pick a change with a clear result

Imagine a model scene with a switch and a door. The door already opens when the switch is pressed. The task is to add a visible indicator above the door when it has opened. Before starting, record the current behavior. Write down which level and asset the agent can edit, which trigger it should reuse and what you expect to see from the normal player camera.

That scope gives the review a simple shape. You can open the asset, inspect the changed properties, play the room and see whether the indicator changes at the right time. If the agent also edits the input mapping, you can identify that as work outside the experiment. The AI-assisted game code review explains how to check generated changes against the surrounding system.

Separate the response from the asset change

An assistant can report that a step finished. Verify the resulting assets in the editor and in source control. Check whether the expected actor or asset changed, whether neighboring assets were saved, and whether references still point at the right objects. Save the small diff and note the engine version and tool configuration used in the test.

The preview announcement also includes known issues. If a tool appears unavailable in a fresh project, consult that section before using a larger test to diagnose it. Record whether the failure was in tool access, applying the edit or playing the result. Each has a different review path.

Run the scene without the assistant

Close the assistant panel, reload the level and test it in play. Then package a smoke build when your target and test setup support it. Check that the indicator starts in the intended state, changes after the player uses the switch and behaves as expected when the level reloads. Verify the same on the original build to keep the comparison grounded.

If you test the experimental Web target separately, document the build requirements and browser you used. A packaged desktop result and a browser result can tell you different things about runtime behavior. Keep the outcome tied to the environment where you observed it.

Save a clean baseline before the first agent action. If the smoke test fails, return to that baseline and reproduce the same change manually on a separate copy. The comparison can reveal whether the issue comes from the preview build, the chosen edit or the way the tool applied it. For an asset-only edit, also reopen the level after restarting the editor. The change should still be available in the saved project without having to reconstruct it from the assistant's transcript.

What should the team leave with?

Record the task, observed diff, playable result and any manual correction needed. For a useful first trial, another team member should be able to read your notes and repeat the same check. If the work is worth keeping, the captured evidence also gives you a basis to decide which parts of the process you want to try again. More practical checks are available in the QORPO development blog.