When a game engine changes, players may notice new tools, smoother performance or smaller updates. They are less likely to see the code that made those changes possible. Epic Games has now offered a striking glimpse behind that curtain: its engineers are using AI tools to develop Unreal Engine, Fortnite and other projects, with AI producing much of the code while people guide the work.
“Yes, we are vibe-coding things,” Marcus Wassmer, Epic’s executive vice president of development, said in a statement to This Week in Videogames. He described engineers as directing AI tools toward the intended architecture, then iterating on the results—not asking a model to build an engine in one go. The distinction matters: Epic is describing AI as part of an engineering process, not a replacement for engineering oversight.
From experiment to everyday development
The disclosure builds on an earlier statement from Wassmer that Epic had opened up broad use of code generation and AI analysis across backend, engine and game-development teams. He said much of the current work is happening in development streams tied to Fortnite, Unreal Editor for Fortnite and Unreal Engine 6. The tools he named include Claude Code and Codex.
That puts AI close to the machinery behind games and the tools used to make them. Unreal Engine is used by developers far beyond Epic, so changes to its performance and workflows can matter well beyond Fortnite. But the company has not presented its AI use as a magic button: Wassmer said engineers are still steering the tools and keeping the engine within limits for memory, performance and scalability.
His case for the approach rests on specific engineering tasks. Epic says AI-assisted work has nearly eliminated a class of non-deterministic “cooking” bugs that had troubled the company for a decade. In game development, cooking is the process of preparing assets and content for a particular build; bugs that behave inconsistently can be especially difficult to track down.
The promised gains—and the unanswered questions
Wassmer also pointed to faster work on shader compression, which he said has helped reduce patch sizes, and a rewrite of a significant part of the Direct3D 12 rendering layer. He estimated that the latter would take one engineer three to six weeks; Epic completed it in about a week, with improved CPU time and fewer bugs, according to his account.
Those are notable claims, but they are Epic’s own assessment of its results. They do not tell us how much code required revision, how the work was reviewed, or whether similar gains will hold across other projects. Nor do they establish that AI-written code is inherently better or worse. The useful question is less whether a tool generated a line of code than whether the finished software is reliable, maintainable and performs as intended.
Epic’s position is that moving cautiously carries its own risk. Wassmer said the company is more concerned about falling behind than about skepticism toward AI, and argued that developers should be candid about how they use these tools. That stance sits alongside a broader debate at Epic: CEO Tim Sweeney has criticized Steam’s AI disclosure requirements, while Epic has itself discussed AI in development and recently backed away from an AI requirement for a game jam.
For players, the immediate consequences are practical rather than abstract. Better engine performance, fewer bugs and smaller patches could make a difference; the process used to achieve them may matter less than the quality of the result. At the same time, Epic’s openness gives developers and players a clearer basis for asking how AI-generated work is checked and what standards it must meet.
Epic is not claiming that an AI can design Unreal Engine 6 on its own. It is saying that its engineers now spend more time directing code-generating tools and less time writing the bulk of the code themselves. The test will be whether that shift produces better tools and games—and whether the human judgment guiding the process remains visible in the results.