Epic Confirms AI Tools Are Writing Unreal Engine Code, Including the D3D12 Layer

Epic Games has confirmed that AI tools are now part of how Unreal Engine gets built. According to a statement Epic gave to This Week in Videogames (reported by GamingBolt and others on October 28), the company is using AI tools to "develop code and features" for Unreal Engine, and coverage says that includes core engine components such as the D3D12 rendering layer.

It is a notable admission from the company behind the most widely used commercial game engine, and one that Epic's CEO, Tim Sweeney, had pushed back on in 2025. Here is what is confirmed, what rests on secondhand numbers, and what builders should take from it. We could not open the original This Week in Videogames piece (it returned a 403 to our fetcher), so details below are attributed to the outlets that did.

TL;DR: confirmed vs reported

What Epic said

Per GamingBolt's account, Epic told the outlet it uses AI to "develop code for Unreal Engine." Sweeney had said in 2025 that Epic was not using AI tools for development, in a discussion tied to Valve's Steam policy on disclosing generative AI in player-facing content. He later wrote, in the quote GamingBolt cites, "Yes, we are now using AI tools to develop code for Unreal Engine." On October 12, 2026, Sweeney described "a huge pivot underway" toward "maximum AI acceleration," according to the same coverage.

Marcus Wassmer, Epic's development lead for Unreal Engine 6, is reported to have said the stated goal is to use AI to "develop code and features" while keeping the engine stable as Epic trims and modernizes it. Unreal Engine 6, previewed in June 2026, is meant to unify Unreal Engine 5 with the Unreal Editor for Fortnite, and GamingBolt says Epic is still testing what works for AI code generation.

A small shape riding a green wave, representing vibe coding in the Unreal Engine pipeline

The D3D12 claim, and how to read it

The headline detail is the D3D12 rendering layer: the part of the engine that talks to Microsoft's Direct3D 12 graphics API on Windows and Xbox. A rendering hardware interface sits near the bottom of the engine, where mistakes show up as crashes, corruption or performance regressions across every game.

Coverage of the statement (via a web search; we could not open the original) says AI helped rewrite a significant portion of that layer, and that Wassmer estimated work taking a single engineer three to six weeks was finished by the team in about a week. That is a plausible and attention-grabbing productivity claim, and it is also an estimate from the company making the announcement. It does not tell us how many review hours, test runs or follow-up fixes surrounded that week, which is where AI-assisted work most often hides its cost.

Epic's other claims, as reported, are of the same kind: AI helped nearly eliminate bugs caused by non-deterministic asset cooking that had persisted for close to a decade, and sped up shader compression. These are specific enough to check against engine release notes over the coming months. None has been audited by a third party yet.

Why it matters: three readings

For engine users and studios

Nothing in the statement changes Unreal licensing, royalties or release dates. What changes is the supply-chain question: if a growing fraction of engine code is AI-drafted, studios that fork or patch the engine, or ship on consoles with strict certification, will want to know how Epic reviews it. Epic says it has deployed automatic stability monitoring for Fortnite on PC, along with automatic root-cause analysis and fix generation for PC crashes in Fortnite and Unreal Engine, per GamingBolt. That is the reassurance Epic is offering: tooling that catches regressions quickly.

For the AI coding debate

Engines are a hard test case. The code is performance-sensitive, heavily shared and old, with large parts written years ago under constraints the original authors may not have documented. An engine team that reports using Claude Code and Codex this way is closer to the "agents on a legacy codebase" scenario than a greenfield app. For background on what that looks like, see our guides to agent harness and async tool calls in Unreal and UEFN's MCP connection for Fortnite agents.

For Epic's public positioning

Epic is running two AI stories at once. It said its new launcher preview was vibe-coded and faster than Steam on cold start, Bloomberg reported on an AI-first Disney shooter with no hand-written code, and a day earlier Epic and Microsoft were accused of hypocrisy for taking down AI-built game clones. The confirmation fills in the base layer: this is not one experimental project but a company-wide pivot. Whether consistent or not is a values question; the facts available say Epic is adopting AI aggressively in engineering while enforcing its IP against others.

Timeline: from denial to pivot

The pattern is a company moving from a defensive position on AI to treating it as a core engineering method in under a year. That is a faster shift than most studios have made, and it explains why the statement drew attention even though the broader industry has been quietly adopting coding agents for months.

How agents typically work on a big, old codebase

A rewrite of a rendering layer is not a prompt that says "port this to D3D12." In practice, teams using Claude Code or Codex on a large C++ codebase tend to do something closer to the following. First they have the agent read the existing module and write a plain-language map of what it does and which other modules call it. Then they split the work into small, independently testable steps. Each step comes with a test or a capture comparison, since rendering changes can be verified by diffing frames. Then an engineer reviews every diff before it merges.

That workflow is consistent with what Wassmer is reported to have said, that engineers no longer write the bulk of the code but are still guiding the tools "to get the architectural result we want." It also suggests where the risk concentrates: not in the generation, but in whether the tests are strong enough to catch plausible-but-wrong changes. Rendering, threading and memory-lifetime code are the places where an agent's mistakes look reasonable in review and fail only under a particular GPU, driver or load. If Epic's claimed week-long turnaround holds, it is a sign that its test and capture infrastructure is doing a lot of the work, which would be the more transferable lesson for other studios than the speed-up itself.

Epic's official documentation already describes its AI features, tools and plugins for Unreal Engine, and Sweeney has separately said, per an IGN interview summarized by Wccftech, that Epic built an MCP server so developers can bring their own agent rather than a proprietary Unreal coding model, and that the tools risk "AI slop" but act as an accelerant for professional creators. The new statement is about Epic's internal use, a different matter from what it ships to customers.

What to ask before you trust AI-written engine code

If you build on Unreal, or on any engine moving this way, practical questions are:

1. Where is the review? Ask whether AI-drafted changes get the same code review and test requirements as hand-written ones. Epic's statement says engineers still guide the tools; it does not describe the review process.
2. What do release notes say? Watch Unreal Engine 5.x and 6 preview notes for regressions in D3D12, cooking and shader compilation, the areas Epic named.
3. Can you pin versions? If you ship on a fixed engine version, you control exposure; if you track the main branch, you absorb the changes.
4. Do your own tools follow the same loop? Many studios are now adopting Claude Code and Codex-style agents for gameplay and tooling code; the engine team's experience is a reference point, not a guarantee.

What is still not known

• The original statement's exact wording, which we could not read directly.
• Whether the "three to six weeks versus one week" comparison includes review and testing time.
• The share of Unreal Engine commits that are AI-drafted.
• Whether Epic will publish incident data, such as bugs traced to AI-written changes.
• Whether console platform holders, who certify engine builds, have asked for disclosure.

We will update this post if Epic publishes more or if the original article becomes readable. For related background on AI in game production, see our coverage of Xbox Labs' AI decompiling experiments.

Related reading

• Epic and Microsoft AI game clone takedowns
• Epic's AI-first Disney shooter
• Epic Games Launcher vibe-coded preview
• Unreal agent harness and async tool calls
• UEFN, Unreal MCP and Fortnite Verse agents
• Claude Code commands reference
• Xbox Labs AI game decompiling

Details are accurate as of October 29, 2026, based on secondary reporting of Epic's statement.