Epic Games Drops Manual Code to Build Marvel Star Wars Shooter with AI

Staff is unsettled, the release date has passed, and Tim Sweeney says the plan is to survive.

Tim Sweeney tried to get ahead of the headlines. The Epic Games CEO posted 24 minutes before a report on the company went live. He said the "AI-lite" era is ending and that Epic has a major pivot underway toward maximum AI acceleration. He said the goal isn't doing the same work with fewer people. It's letting everyone at Epic build more and better things and passing those benefits to every team using Unreal Engine.

The report tells it differently. Citing people inside the company, it says the pivot mandates AI-first development on Epic's next game, an extraction shooter already stuck in development hell. Leadership reportedly thinks building all the art and code with AI can rescue it.

For now, the game is delayed with no date, and many staff don't share Sweeney's confidence. The game isn't a Fortnite project; it's a standalone extraction shooter and a Marvel-Star Wars crossover. That fits Epic, since Star Wars, Marvel, and other Disney properties have done well in Fortnite, before and after Disney invested $1.5 billion in Epic in 2024.

To be fair, a lot of Epic's work, animation for example, is well done, even if many players see slop. Crossovers like this are a long tradition, and it isn't the only Marvel and Star Wars team-up Disney has planned for next year's anniversary of the original Star Wars, since a comic is coming too. Decades of art, models, and stories make an AI-driven approach seem feasible on paper, even if something fresh is harder.

The pitch is closer to "remember this, remember that, here they are together." The theories behind AI-first development are now orders of magnitude ahead. The report says a game director posted in an internal chat that there would be "no more hand-written code" and that all art must be "generated with full AI workflows."

On the code side, Anthropic's Claude is reportedly the model of choice, and programmers are expected to use it.

Forcing a tool on people changes their relationship with it, no matter how good it is. You can assume they're running something like Claude Opus, which Anthropic pitches as strong at coding and at working autonomously for long stretches.

That's an agentic workflow: you hand it a scoped task, and it plans the work, writes the code, delegates to sub-agents, finds and fixes bugs, and even opens other software when needed. Tests across engineering and knowledge work suggest they mostly deliver, which is surprising.

None of that means an AGI god is arriving. It means software engineering is improving fast but unevenly across task types. Advanced C++ in Unreal Engine 5 is a long way from the kind of Claude demo that goes viral. The order behind AI-first development isn't limited to code. Code has a verifiable outcome you can hill-climb toward, which adds reliability. Art is different.

Details are thin, but treating AI as a machine that spits out finished assets would be a bad idea even for Epic, and the report suggests that isn't the plan. Humans will manage and guide the work. That's probably how Epic plans to handle copyright, since the US Copyright Office says work made by AI without enough human control can't be copyrighted.

Leadership believes AI for code and art will speed things up, and that developers steering agents with a human creative vision should keep them in the clear on copyright.

Insiders say developers across Epic had been trying these tools on their own and found good uses. Epic Games clearly knew the report would go down badly with players, which is likely why Sweeney posted first. The company didn't deny anything, saying it is embracing AI-accelerated development led by its developers, whom it calls the visionaries and creative drivers behind everything it does.

If you only listen to Tim Sweeney, you'd think nothing can go wrong, but insiders say the developers were stunned on a call earlier this month. Some joked about asking Claude to fix its own mistakes. Others hoped the directive wasn't serious or that Epic would reverse it, arguing it wouldn't work as well as hoped.

The bigger issue is that AI-first development is being imposed on everyone on the game, not just developers who found value experimenting on their own. Epic is overhauling its entire pipeline and telling everyone to use these tools regardless of how well they fit or what they produce, without knowing whether they'll help.

Happy developers are unlikely to leak, but that doesn't change how the staff in the report feel, and they doubt AI can fix the game's biggest problems. AI-first development won't fix those human problems. Sources say progress on the Disney game has been slower than expected, partly because Epic underestimated the staff and time needed, which is ironic after it laid off 1,000 employees in March.

That reads as a planning failure by leadership, not a delivery failure by staff. The timeline agrees. The game was meant to launch this autumn, slipped to next March, and now has no date. The causes sound typical for Epic: slow decisions, conflicting directions, constant pivots, and endless changes. Fortnite itself only became a battle royale after the original game launched.

Epic also hasn't made a new game since Fortnite, and shipping only seasonal updates erodes pipeline expertise.

Slow, conflicting decisions look like the main hurdle, and AI-generated code or art won't fix that. Poor pre-production planning and a lack of creative vision are the biggest problems in game development today. AI-first development might help staff react faster when leadership changes its mind. Optimistically, that means faster iteration and ironing out design issues before you commit big money. 

Pessimistically, weak creative vision is a human problem, and a code generator only spins your wheels faster.
Viewed on its own, the Marvel and Star Wars project looks like a Hail Mary, but there's more to it. Tim Sweeney already explained the timing. In his post, he said he wants to bring these benefits to all Unreal teams.

When asked whether it would speed up Epic going public, he said the aim is more modest: survive the AI revolution, help other developers do the same, and come out making better code and content faster. The key part of that reply is the mention of other developers. Earlier this summer, coverage of Epic's AI voice tools and early talk of Unreal Engine 6 built around AI-native development.

Epic licenses Unreal Engine across the industry, so if it builds a successful game with an AI workflow, builds the engine around it and develops the tooling, it can sell that to anyone. Some teams have already tried AI-first development, though.

Luke Dicken, a former head of AI at Take-Two who now runs his own consultancy, challenged the idea that this is the first such project, and the reporter clarified he meant the first public attempt by a major studio. Dicken said several efforts exist that nobody talks about because they're polarizing even internally, then fail and get quietly buried. The polarizing part is true here.

Epic hopes to repeat its habit of using licensed properties to market new technology, as with the AI Darth Vader voice, but it must prove AI-first development delivers. 

The models have improved, but a tool's capability in isolation differs from how it performs in a live pipeline.
This is a big bet by Epic Games leadership. With the project stuck, they seem to feel they have little to lose, so making a licensed title the first public test of AI-first development looks like a gamble.

Sales aren't the point. If it works, the real product is telemetry and session traces from coders on a live commercial project, which feed an engine built around a new development model that Epic will pitch as the AAA standard. Epic and its partner wanted a launch alongside the next Avengers film or Star Wars milestone. Delaying it indefinitely signals they don't believe staff can hit those dates, so they've stopped asking. 

Even if AI-first development speeds up engineering, code is only one factor. Agentic coding can build systems quickly, but the product still has to be compelling, and the people steering it made the slow, contradictory decisions the insiders describe. If you're hoping for a Star Wars-and-Marvel extraction shooter, keep your expectations in check.