Epic x Disney Game Delayed, Shifts Fully to AI-First Development: What ‘No More Hand-Written Code’ Reveals About Big-Studio AI Workflow Overhauls and Developer Pain#

Hello everyone, Wubian here. The story most worth watching in games this week was not a sales chart but an internal directive: according to Bloomberg reporter Jason Schreier’s October 13 exclusive, Epic Games has delayed an unannounced shooter developed with Walt Disney and asked the project’s developers to move fully to an “AI first” production model — using AI tools such as Anthropic’s Claude to generate code and art, with human developers shifting to managing and directing what the AI produces.

Once the news landed, one sentence a project director wrote in an internal message seemed destined to become the most quoted line of this year’s industry coverage:

“No more hand-written code.”

No more hand-written code. Behind those five English words sits a flagship studio rewriting what development means. After years of watching AI and game ecosystems, I have seen plenty of “AI-assisted” pilots and broken down plenty of tool benchmarks, but this is the first time a top publisher has put humans-not-writing-code in writing as a requirement on a flagship project. In this piece we break the story into four layers: the organizational meaning, the return-on-investment math, the data behind the industry’s tensions, and the demonstration effect and player trust.

1. First, the facts: a game that changed how it gets made#

As always, the confirmed information goes on a timeline first, judgment later. The facts below follow the Bloomberg report and cross-checks across other outlets:

The most striking stretch of that timeline is the six months between March and October: in the March layoff announcement, nothing to do with AI was still the officially repeated line; by October’s internal directive, AI had become a core variable written into production requirements. In half a year, the same company’s public positioning of AI completed a one-eighty — that is the key to the whole story, and every layer below loops back to it.

Now the game itself. It is a shooter built around Marvel and Star Wars character crossovers, the first title of the $1.5 billion partnership; earlier Bloomberg reporting described it as an extraction shooter, and in the industry it was discussed as a lookalike for Embark’s ARC Raiders — to be clear, the extraction label comes from media accounts and was never officially confirmed. The game was originally set for March 2027; it is now delayed with no new date, no announced platforms, and not even an official title.

That extraction label deserves a word more. The genre has been brutally competitive for two years, ARC Raiders has just posted genuinely strong numbers worldwide, and the window for a major publisher to follow into the same theme was always going to be short — a March 2027 slot in this lane meant Epic arriving after others had already taken their positions. That context gives behind schedule a heavier weight: what slipped was not just the schedule but the genre window. That said, this is background observation based only on public information; Epic has never tied the delay to genre competition, and we will not over-extrapolate.

The direct cause of the delay, per Bloomberg, was that the project ran behind schedule: Epic underestimated the headcount and development time it needed, on top of considerations around the release plan. AI first was not the cause of the delay; it was the response and the transformation plan after it. Many headlines have already scrambled that causal order. We should not follow them.

Separating what has been officially confirmed from what the outside world still does not know helps map the information boundary of this report:

This table matters because it draws the boundary of certainty around the story. All the extrapolations about Epic scrapping hand-written code across the board, or AI first being a prelude to another round of layoffs, currently cross that line — what the reporting supports is the delay of one game and a switch of production model, plus one CEO confirming a pivot.

2. The organizational meaning of AI first: developers become supervisors#

The most interesting part of this story is not whether AI gets used but how the work gets organized. As the report describes it, the production model developers are being asked to run is: AI tools such as Claude generate the code and art, and human developers go from the people who write code to the people who manage and direct the AI’s output.

How fundamental is that shift? Here is a side-by-side of the traditional pipeline and the AI first pipeline:

Start with the most uncomfortable row in that table: quality control. Bloomberg’s reporting makes no mention of acceptance criteria or accountability mechanisms to match full AI generation — when something breaks, whose fault is it; who stops style drift; how Marvel and Star Wars IP material clears review — none of those questions has an answer yet.

The reaction on the ground confirms the unease. At that video call in early October, some developers were stunned by the new directive: one joked about whether Claude should be made to fix the errors it produced itself; another hoped this was not a serious request and that the company would come around; a third judged flatly that the plan would not work out the way Epic expects. Behind the joke sits a very specific career anxiety — when the generator and the acceptor collapse into the same tool, who owns the failure: the person directing the AI, or the model? Inside this company, that is no longer a philosophy question. It is a performance question.

Laying out the four frontline reactions and the questions each points to says more about the temperature inside the organization than any single quote:

Read together, none of the four reactions — shock, the joke, wishful thinking, doom-saying — is about how to actually run an AI workflow. The frontline is still at the stage of whether to accept this, while management has already moved on to execution; between the two sits an organizational fault line. Big transformations historically succeed when the doubts of the people executing are converted into a path to participate, and what Epic has offered so far is one internal directive and one CEO pep talk.

Now Epic’s official line, worth parsing sentence by sentence. Communications director Jake Jones told Bloomberg that the company is embracing AI to accelerate development, led by Epic’s developers, who remain the visionaries and creative drivers of everything it makes. The most carefully chosen words in that statement are led by our developers — the intended signal is that people have not been removed, that people are still in the driver’s seat. But set it against the internal directive and a subtle mismatch appears: the statement calls people the leaders, while the directive moves them from producers to supervisors — and the work in between, the part that used to be called doing, has been quietly handed to the model.

Sweeney was blunter in his October 12 post on X: this is not about doing the same work with fewer people, but about everyone building more and better things, to the benefit of every team that uses Unreal Engine. The sentence has two edges. Inward, it is a mobilization order — everyone doing more translates into performance language as a higher per-head output target. Outward, it reassures engine licensees — benefiting every Unreal Engine team makes clear the workflow will not stay locked inside Epic. And to critics he offered a strikingly cold response: many people in the room want to go back to 2016, but that is no longer possible; as the company supplying tools and technology to most of the industry, Epic’s job is to design for the present and the near future. That is no longer possible effectively ends the internal debate: this is not a consultation, it is a decision already made and already in motion.

Put the two accounts side by side and you get a classic collision of transformation narratives: in the official story, developers are upgraded from executors to visionaries; in the frontline experience, they are downgraded from producers to acceptors. The same role change reads as empowerment looking up and as the prelude to replacement looking down — which reading comes true depends on how Epic assigns acceptance responsibility, and on whether it can keep the promise of not doing the same work with fewer people.

3. The return-on-investment math: $1.5 billion, 1,000 people, and one delay#

Layer two goes back to the money. Read the timeline from 2024 to now as one continuous line and it is a textbook big-publisher ROI pressure chain:

1. In 2024, Disney invests $1.5 billion in Epic for an equity stake, with a multi-game partnership attached; this Marvel x Star Wars crossover shooter is the partnership’s first title. For Disney it is a strategic bet on platform-style virtual worlds; for Epic it is money with delivery expectations attached.
2. In March 2026, Epic lays off about 1,000 people. The official attribution was slowing Fortnite sales, with some outlets adding weak console sales and the litigation with Apple and Google as context; Epic made a point of saying the layoffs had nothing to do with AI.
3. In October 2026, the project is delayed and production switches to AI first, and the efficiency narrative from management formally takes over from the cost narrative.

Line the three moments up and one observation is hard to avoid: the layoffs that had nothing to do with AI and the all-in AI first transformation, six months apart, at the same company under the same business pressure. Epic has never said AI first will bring further layoffs, and that must be stated plainly; it is equally true, though, that the reporting line about running behind schedule and underestimating headcount and time is itself telling you something — under a traditional staffing model, the math on this game no longer works. A $1.5 billion investment, a four-digit headcount, a promised 2027 slot: when those three collide, AI acceleration becomes the only lever left that can hold all three at once.

Then take the market signal each moment sent on its own:

• $1.5 billion (2024): bought equity plus a multi-game path; Disney was betting on the virtual world platform narrative, and the delivery rhythm carries the patience of capital markets.
• About 1,000 layoffs (March 2026): the official attribution was slowing Fortnite sales, with some outlets adding weak console sales and the Apple and Google lawsuits — not one of the three attributions mentioned AI, a line that was deliberately maintained at the time.
• The skin concept art controversy (June 2026): the first time the player community filed Epic under AI-generated content; the tuition on public opinion got paid early.
• The AI first directive (October 2026): the cost narrative formally switches to the efficiency narrative, and the carefully kept rhetorical distance between layoffs and AI effectively drops to zero in one move.

Read the four steps in sequence and the distance from nothing to do with AI to AI first is six months. Business pressure did not ease in that time; what changed was the toolkit — in March 2026 the available move was cutting headcount, in October 2026 it is swapping the production function. That pivot point may be the single most worth-recording item in this whole story for other publishers.

Sweeney has also supplied his own efficiency numbers, worth recording separately (these figures come via a single source relayed by Gikutaku, so caveat emptor): by his account, the most technical parts of Unreal Engine development are seeing efficiency gains of about 20 percent, web development and similar work up to 5x, and he predicts that within a year few tasks will be below 3x their January 2026 level. The 20 percent is the honest number — it says the AI gain on core technical work is incremental improvement, not an order-of-magnitude jump. The 5x on web development is a comparatively mature AI use case, and holding it against game development is a slight category error. The 3x prediction reads more like an internal mobilization order, setting next year’s expectations in advance. Read the structure of the numbers and a pattern emerges: the further from core game engineering, the bigger the AI gain — and that gradient itself is a quiet correction to the aggressive narrative of AI replacing core development wholesale.

Another telling detail is Disney’s posture. Bloomberg asked Disney for comment; a spokesperson did not immediately respond. And per a report relayed by Chinese-language media (single source), Disney has also recently denied having any intention to acquire Epic outright. We broke down the calibrated wording of Disney’s response to the acquisition rumors last month, and the read then was a denial of the present, not of the future — which looks even more patient in hindsight: the $1.5 billion did not buy a game, it bought a partnership that can afford to wait. Even with the first title delayed and switching production models, Disney has yet to apply public pressure. That silence is itself the investor’s observation window on the transformation.

Viewed as communication strategy, Disney’s silence and Epic’s loudness make a complementary pair: Epic needs to prove its resolve to the industry, so Sweeney posts personally and answers critics; Disney needs to protect the public environment around its investment, so it declines to engage. Neither side has miscast itself — the people truly squeezed are in the middle: project team members who have seen layoffs, seen the AI directive, and heard no promise at all about what happens to the people. For this company, external PR is far easier than internal communication, and the reaction in that video call already showed it.

4. The scissors gap in the data: studios speed up, developers cool down#

Layer three pulls the camera back from Epic to the industry’s thermometer. Two surveys in one table and the tension is plain:

In the same window, the share of people using the tools is growing, and the share put off by them is growing faster — negative sentiment went from three in ten to more than half in a year. That is the scissors gap: the adoption curve on the studio side and the negative-sentiment curve on the workforce side are steepening in opposite directions at the same time.

One caveat when reading these surveys: the samples differ. GDC’s sample is about 2,300 industry professionals across game-industry jobs in the broad sense; the Google Cloud 87 percent is a number Gupta relayed, with no sample composition disclosed alongside the report. The former measures using AI tools at work, the latter using AI agents, and the definitions are not the same width — putting 36% and 87% side by side directly would distort. But once you allow for that, the two surveys actually agree more: however you count, AI penetration in development workflows is at or past the halfway mark. The argument was never about whether this is happening; it is about on what terms.

There is a more alarming reading of the scissors gap. A 36% usage rate means most of GDC’s sample — over six in ten — are not yet using AI at work, and negative sentiment is already at 52%. In other words, the main force behind the negativity is not people who tried it and were disappointed, but a large body of wait-and-see watchers who have not picked the tools up and have already seen which way the wind blows. Their anxiety comes not from experience but from expectation: when a company like Epic writes no more hand-written code into an internal directive, the message for professionals not yet using AI is that the time left for not using it is running out. Negative sentiment among watchers peaking before adoption is a pattern seen in earlier technology transitions; this time, because AI’s capability boundary overlaps creative work more closely, the intensity may run higher.

Curiously, the loudest defense of Epic’s approach comes from a veteran insider. Former Fortnite producer Aakash Gupta posted a long thread on X that breaks into four points:

1. The speed story: Fortnite’s battle royale mode was built in two months back in the day — using Epic’s own history to show that small teams plus a ruthlessly focused process have always been this company’s core skill;
2. The organizational story: at Epic, a transformation means the whole company turns that same week — he reads AI first as a routine flex of Epic’s execution culture, not a radical experiment;
3. The adoption story: he cites the Google Cloud survey that about nine in ten (87%) developers already use AI agents at work — everyone is already doing this is the foundation of his whole argument;
4. The verdict: the industry’s AI fight is mostly for show — the real argument is not whether to use AI but who admits it and who digs in.

Of the four, the first three are fact and experience and the fourth is a value judgment, and the backlash has concentrated on the fourth. From a practitioner’s seat, everyone is already using it is a statement of fact; from labor’s seat, it is precisely the problem — once using AI becomes the default, bargaining power flows back to capital, and the irreplaceability of the individual developer gets repriced. The 52% negative share in the GDC survey reflects exactly that structural anxiety: the more ubiquitous the tool, the more passive the individual.

But he adds a more granular observation too: by his account, Epic has already connected AI models including Anthropic’s Claude and Google’s Gemini to Unreal Engine, with more integrations under way. In other words, while the argument rages in public, the pipes are already in the editor.

It is worth flagging Gupta’s position, though: he is a former producer at the company, now active in the industry as an AI evangelist, and his optimism is naturally in phase with Epic’s current line. Read him as the best annotation for why Epic is doing this and he is useful; read him as a neutral referee of the transformation and he distorts — like his mostly for show line about the industry debate, it is an observation with a standpoint: valuable, but it needs calibration before use.

The industry has samples at the other pole too. According to the same report, Level-5 president Akihiro Hino says 80 to 90 percent of the company’s code is written by AI and then reviewed by programmers, and Shift Up, the studio behind Stellar Blade, says 150 people now produce the output that used to take 1,000 to 2,000. Put Level-5, Shift Up, and Epic together and the difference is not whether to embrace AI but where the line sits: the first two describe a safety-net structure of AI produces, humans review, while Epic is explicitly requiring that people be moved out of production altogether — which is exactly what makes AI first far more radical than AI assisted.

5. The demonstration effect: an engine maker flips the default to on#

Layer four is the one I think actually reshapes the industry: Epic is not just a game company; it owns Unreal Engine. That means the blast radius of this transformation is far larger than one delayed game.

The chain runs like this: per Gikutaku (single source), Unreal Engine 5.8 already integrates Claude and Gemini models, and the engine’s next version will ship with a protocol that makes connecting AI models simpler. List the diffusion in steps and you see why it is close to irreversible:

1. Step one, validate on owned projects: Epic trials the full AI workflow on its own Disney project; the successes and failures alike settle into engineering practice;
2. Step two, capability ships in the engine: model integration is already in version 5.8, and the next version adds a protocol that simplifies connections — capability moves from an Epic internal practice to an engine standard part;
3. Step three, licensees get it by default: millions of Unreal developers need not buy anything; upgrading the version delivers it, and AI goes from value-added service to built-in feature of software already paid for (which is exactly Gupta’s point);
4. Step four, the industry standard shifts: when the engine ships with an AI channel, hiring requirements, outsourcing quotes, and development-cycle budgets all get recalculated — studios that do not use it have to explain why, and studios that do never do.

The industry consequences deserve their own list:

• Adoption no longer requires a decision. Whether a studio used AI used to be a management choice; going forward it is a checkbox in the editor — and not ticking it is what needs a reason.
• The cost of resistance goes up. A studio wary of AI-generated content will find its cost structure systematically disadvantaged against the shop next door, because the engine it uses has raised the competitor’s efficiency baseline wholesale.
• Passive catch-up at the industry’s tail. Former Square Enix executive Jacob Novak put it bluntly: not keeping up means being left behind. For large numbers of small and mid-size teams, this is not a multiple-choice question; it is a survival question.
• The controversy gets diluted by engineering. When AI capability arrives as protocols and version updates, the discussion shifts from whether it should be used to how — and the latter has always been an agenda item for engineering meetings, not ethics ones.

This is why I call the story a turning point: Epic walks on two legs, Fortnite and engine licensing, and the engine leg reaches the whole industry. An engine maker fully validating an AI workflow on its own flagship project means the results, good or bad, get written into the daily toolchain of millions of developers as protocols and version updates. The other big publishers — Activision, Ubisoft, Tencent — will not wait long.

6. Player trust: June’s precedent and the unresolved accountability question#

The last layer returns to the people who buy the games. Epic has already run one rehearsal on player trust this June: it published a video of skin concept art made with generative AI, touting AI-accelerated design with human correction, and the comment section turned near-uniformly hostile, with players demanding a disclaimer on AI-generated content. Set June against now and the change in scale is obvious:

Back then it was concept art assistance; this time, per the internal account, all art goes through the full AI workflow — a different order of magnitude. And what players did not get in June — labeling, disclaimers, a right-to-know mechanism — shows no public sign of arriving this time either.

Break the trust risks down by stage:

1. Art quality and style consistency. Marvel and Star Wars are two of the most detail-obsessed fandoms on earth; any drift in a character’s look will be frame-checked. Whether art from a full AI workflow can reliably pass that test has no precedent — in June the argument was over whether the style looked cheap; this time it will be over character fidelity, where the margin for error is far thinner.
2. The hidden cost of code quality. The joke about having Claude fix its own errors spread among developers because it points at real technical debt: model-generated bugs fixed by the model, an error loop that can close on itself — who certifies that it is fixed? In an always-online shooter carrying two flagship IP licenses, that class of error bills directly to service stability.
3. Disclosure obligations. The core demand in June was the right to know. When art and code are both products of an AI workflow, how should marketing materials, copyright labels, and asset-provenance statements be marked? The industry has no binding standard, and Epic has not published one either.
4. An accountability vacuum. If AI-generated assets trip over copyright or likeness rights, Epic and Disney answer outward, but the path of internal accountability has never been explained — the traditional chain of whoever signed off owns it has no equivalent in an AI workflow.

These four risks share one feature: none of them is about whether AI does the work well; all of them are about whose desk the problem lands on afterwards. Technical problems ease as models iterate; accountability problems deepen as the process hardens — and the latter is the true foundation of player trust.

For a company holding a crossover of two top-tier IP families, a failure at any of these points damages not just one game but the very proposition that an AI workflow can carry flagship content. Flip it around, though: if this works, Epic has blazed a trail from pilot to mass production for the entire industry — which is the real reason every side is watching this project.

One line each on what this means for three kinds of readers. If you are a developer: the thing to watch is not whether to learn AI tools but how supervisory responsibility gets defined — the 52% negative share in the GDC survey is, at bottom, a protest against that definition’s absence. If you are a player: from June’s concept art to this full workflow, the right-to-know demand is still stuck at the raising stage; next time you buy anything with AI’s fingerprints on it, it is worth checking whether the labels are complete. If you follow the industry: two signals deserve sustained tracking — the details of the AI protocol built into Unreal Engine’s next version, and whether the game’s new release date comes with any AI content disclosure. One sets the industry’s diffusion speed; the other sets the baseline of player trust.

7. Closing: craft in the age of supervisors#

Having come this far, I want to end on the people. As a longtime observer of AI deployment, I do not actually doubt the arithmetic of AI first — the $1.5 billion from 2024 needs a return, the 2027 slot needs someone to make good on it, and the efficiency lever is genuinely the most convenient one to pull. Sweeney’s line about wanting to go back to 2016 but not being able to is, on industrial logic, hard to fault.

What made me stop was two details from that video call: someone joked about having Claude fix its own mistakes, and someone hoped none of it was serious. The smartest engineers at a major company, at the turning point of their own project, left with nothing but jokes and wishful thinking — and that tells us more accurately than any survey what the omitted half of the transformation story looks like. Development has never been just producing code and art; it is a whole craft of judging what is right. When production is handed to the model, does that craft rise into the art of supervision, or quietly shrink into the habit of nodding? That is the most expensive chip on the table in this gamble.

I have an old habit from analyzing online game economies: to tell whether a system is healthy, ignore what the launch event says and watch how its participants carry themselves. Same here — the shock and the jokes in that Bloomberg-reported call are how participants carry themselves. In a healthy transformation, the frontline conversation should be technical: how do we use this. When the conversation is stuck on whether this is serious, the organization has not yet answered what happens to the people. Until that is answered, however pretty the efficiency numbers, the pain has merely been moved from the spreadsheet to the human heart.

Disney’s silence continues, Sweeney’s mobilization goes on, and the game without a name will show up someday, somewhere, in some form. One thing is certain: as of this week, page one of the job description for game developer has been rewritten. Whether turning that page brings a leap in output, or an era in which everyone reviews and no one answers for the work, will only be settled the day that game ships.

— Wubian, NTGame Editorial