On 12 October 2026, Bloomberg revealed an instruction given to the team developing Epic Games’ next game, a collaboration with Disney bringing together Marvel and Star Wars characters: “no more hand-written code”. The game, due out in March, has been pushed back with no new date, and its production is becoming “AI first”: artificial intelligence first, developers second, to guide it and review what it produces. That same evening, Tim Sweeney, founder and head of the company behind Fortnite, confirmed on X a “huge pivot” of the whole company towards AI.
Within a week, Epic released a new launcher it says it “vibe coded”, its boss put figures on the expected gains, from 20% to five times faster, then clashed publicly with a respected figure in the industry. We separated what is official from what leaked, then set Tim Sweeney’s figures against the two most solid independent measurements published in 2026 of the real productivity of developers who code with AI. The gain exists. It is a long way from the promises, and its cost shows up exactly where Epic is now placing its developers: in code review.
What Epic Games announced, and what only leaked
The story is going around as “Epic bans its developers from coding”. It blends three levels of information that need to be kept apart.
What Bloomberg reports. According to reporter Jason Schreier, who relies on people close to the project and on an internal message, Epic has delayed its Disney game and asked the team to use more AI to finish it. That message, from the game’s director, reportedly announced “no more hand-written code”, and art now “generated with full AI workflows”. Programmers would guide tools such as Claude, Anthropic’s model, and review the result. The announcement, made on a video call, left some developers “stunned”, according to the people interviewed.
What Epic confirms. The company’s spokesperson does not repeat the instruction. The statement reads: “We’re embracing AI-accelerated development, led by Epic developers who are the visionaries and creative drivers for everything we do.”
What Tim Sweeney says in public. On X, on 12 October: “We have a huge pivot underway here at Epic, moving developers around to retool for maximum AI acceleration. The aim isn’t to do the same stuff with fewer people, but to enable everyone here to build MORE and BETTER stuff, and to bring the benefits to all teams using Unreal.”
| Date | Who | What | Status |
|---|---|---|---|
| 24 March 2026 | Tim Sweeney, note to employees | More than 1,000 layoffs, "not related to AI" | Official |
| 12 October | Bloomberg | Disney game delayed, "AI first" production, "no more hand-written code" | Reported (anonymous sources, internal message) |
| 12 October | Tim Sweeney, X | "Huge pivot" towards maximum AI acceleration | Official |
| 12 and 13 October | Tim Sweeney, replies on X | From 20% to five times faster today, at least three times within a year | Official, no published method |
| 14 October | Epic Games Store | New "vibe coded" launcher, in Preview | Official |
| 18 October | Tim Sweeney, X | The Unreal Engine 1 days: 30% hard problems, 70% plain typing | Official |
| 18 October | Tim Sweeney and Kenney, X | Public exchange, ending with "Up yours!" | Official |
Sources: Bloomberg, Epic Games, Tim Sweeney’s X account, Wccftech for the replies of 12 and 13 October.
One point gets lost in most of the coverage: according to Bloomberg, the “no more hand-written code” instruction is aimed at the Disney game’s team. Nothing in Epic’s public statements indicates that it applies as such to Fortnite or to Unreal Engine. What Tim Sweeney has confirmed is a general direction, and it covers the whole company.
A Disney game running late, handed to AI to get it finished
The project is not just any game. On 7 February 2024, Disney announced it would invest $1.5 billion to take a stake in Epic Games, and the two groups promised to build together a games and entertainment “universe” connected to Fortnite. Players were to be able to play, watch, shop and engage with characters from Disney, Pixar, Marvel, Star Wars and Avatar, all of it built on Unreal Engine, Epic’s game engine.
The game revealed by Bloomberg is the showcase for that partnership. It was due out in March; it no longer has a date. And the order of events matters. According to the people Bloomberg spoke to, development was already progressing more slowly than expected, partly because Epic had underestimated how many people it would need. AI is therefore not presented as the cause of the delay. It is presented as the way to make up for it, on the company’s most exposed project.
Vibe coding: what it is, and why the word is a problem at Epic
The term has an author and a date. In early February 2025, Andrej Karpathy, a co-founder of OpenAI and former head of AI at Tesla, described on X “a new kind of coding” called vibe coding, where you “fully give in to the vibes, embrace exponentials, and forget that the code even exists”. The post spells out the practice: always clicking “Accept All”, no longer reading the changes the AI proposes and, when a bug won’t go away, working around it or asking for random changes until it disappears. The limit is set in the same post: it is “not too bad for throwaway weekend projects”.
The word has since broadened. It now covers any programming done by giving an AI instructions in plain language, including on consumer platforms such as Base44, which we tested. Tim Sweeney gives a definition that differs from Karpathy’s: in a post of 18 October, vibe coding means prompting the AI “and reviewing the results”.
That is the whole question. In the original definition, nobody reads the code. In the model Bloomberg describes for the Disney game, developers review everything. These are two opposite practices under the same word, and Epic uses that word for a commercial product used by millions of players. Yet it is precisely on review that the success of the bet depends, and that is where the independent measurements are least reassuring.
The Epic Games Launcher, the first openly “vibe coded” product
Two days after Bloomberg’s report, the Epic Games Store published the announcement of its new launcher, the application that gives access to the games bought on the store. The announcement opens: “Programming is hard. So, instead of doing that, we vibe coded a new Epic Games Launcher that’s way faster and more responsive than the current one, and has a new look.”
The launcher is offered in Preview for Windows, macOS and, for the first time, Linux. Epic warns that this first release “is light on features” and asks users to “think of it as a speedboat and not a yacht”. The announcement was later updated with a note: the launcher is in testing and “with any beta there could be issues”. Fortnite will only come to Linux once the Easy Anti-Cheat system has been rebuilt.
Epic has published neither the share of the code written by AI, nor the size of the team, nor how long the work took. The only vibe coded product it has shown is therefore a Preview application, deliberately limited and released with a warning. It is not yet proof either way. Its critics are already using it, though: on 18 October, Kenney, the creator of a large library of free game assets, replied to Tim Sweeney that the work should be allowed to speak for itself and that, given the backlash to the “vibecoded” launcher, it did not yet.
Tim Sweeney’s figures, and what they leave out
Epic’s boss did not stop at announcing a pivot. Asked on X how much acceleration to expect, Tim Sweeney replied, according to the exchanges reproduced by Wccftech: “Right now, it’s highly variable, from 20% in some areas (like the most technical areas of UE dev) to 5x for other areas (like web dev). By this time next year, I expect very little will be less than 3x of January 2026 efficiency.”
The Epic Games boss also named the main expected benefit: “much faster time from having an idea to playtesting it”. According to Tim Sweeney, most recent big-budget failures came from game systems that went through too much of their development without being playtested.
On 18 October, Tim Sweeney set out the reasoning. Back when Unreal Engine 1 was being written, in the 1990s, around 30% of that time went to hard problems, and 70% was “just typing and refactoring” large bodies of code. So handing that 70% to AI always seemed logical, once it could do it “with a similar quality standard as we could, which has never been perfection”. The surprise, according to the post, is the emergence of AI as “a sort of insta-Carmack”, a reference to John Carmack, the programmer behind Doom and Quake, able to solve hard technical problems. The day before, the same account had promised “a new golden age of gaming”.
These figures are an executive’s estimates. Epic has published no method, no scope and no data. They do have one quality that is rare in this debate: they are precise, and therefore testable. A “5x” gain means a task takes 80% less time. A “3x” gain means 67% less time.
What independent studies measure about developer speed
The most rigorous measurement available comes from METR, an independent AI evaluation lab. Its method is the controlled trial: experienced developers work on real open source projects they know, and each task is randomly assigned to be done with or without AI. The times are then compared.
| METR study | Effect of AI on completion time | Confidence interval |
|---|---|---|
| Early 2025 (February to June), experienced open source developers | 19% more time | from 2% to 39% more time |
| Late 2025 (from August), developers already followed in 2025 | 18% less time | from 38% less time to 9% more time |
| Late 2025, newly recruited developers | 4% less time | from 15% less time to 9% more time |
Source: METR, 24 February 2026.
In early 2025, then, AI slowed these developers down. By late 2025, it speeds them up slightly, but in both groups the confidence interval includes a slowdown: the study cannot rule out no effect at all. METR is the first to say so, and goes further: it considers these new results unreliable, and probably too low. More and more developers refuse to take part because they no longer want to work without AI, and among the participants surveyed, 30% to 50% say they hold back some tasks for the same reason. METR also cut the participants’ pay, from $150 to $50 an hour, and cannot properly measure the time of those who run several AI agents at once.
Its conclusion is therefore cautious in both directions: it is “likely” that developers are more sped up in early 2026 than in 2025, but its data is “only very weak evidence for the size of this increase”. And it adds a warning that applies directly to the kind of figures Tim Sweeney puts forward: some developers self-report very high speedups, but, as its first study showed, those estimates “can be quite unreliable”.
| What Tim Sweeney claims | What the independent measurements say |
|---|---|
| 20% gain on the most technical parts of Unreal Engine | In the same range as METR's best estimate (18% less time, about 22% faster), but without statistical certainty |
| Up to five times faster in web development | Neither of the two independent sources cited here comes near that order of magnitude; METR calls self-reported gains "quite unreliable" |
| Almost nothing below three times January 2026 efficiency within a year | Faros: 33.7% more tasks per developer between low and high AI adoption, with 54% more bugs |
| AI works to a quality standard "similar" to ours | Faros: incidents relative to the number of pull requests more than tripled |
Sources: Wccftech and Tim Sweeney’s X account for the statements, METR and Faros AI for the measurements.
The real bottleneck: reviewing the code AI writes
METR’s study measures how fast a developer completes a task. It says nothing about what happens next, when that code has to be reviewed, merged and maintained. That is the subject of the 2026 report from Faros AI, a company that sells tools for measuring the activity of development teams. Faros analysed two years of engineering data from 22,000 developers and more than 4,000 teams, in companies using Claude Code, Cursor, GitHub Copilot or autonomous agents. For each team, it compares its two quarters of lowest AI adoption with its two quarters of highest adoption.
| Indicator | Change between low and high AI adoption |
|---|---|
| Tasks completed per developer | +33.7% |
| Large pieces of work ("epics") completed per developer | +66.2% |
| Pull requests merged per developer | +16.2% |
| Deployments per week | −11% (measured on 10% of the data) |
| Bugs per developer | +54% |
| Incidents relative to the number of pull requests | more than tripled |
| Median time spent in review | +441.5%, about five times more |
| Pull requests merged without any review | +31.3% |
| Code deleted relative to code added | +861% |
Source: Faros AI, AI Engineering Report 2026, analysis as of March 2026.
Throughput really does go up: a third more tasks, two thirds more large pieces of work. But Faros sums up its own finding in one line: “More code. Not production-ready code.” Pull requests are bigger, arrive faster, and review cannot keep up: median review time is multiplied by five, and the number of changes merged without being reviewed by anyone rises by 31%. According to the report, 60% of AI-generated code is now being accepted into codebases.
Two caveats apply. Faros sells measurement tools, and presents its results as correlations, not causes. Its data comes from software companies in general, not from video game studios. But it points precisely at the step on which Epic’s model rests. If the Disney game’s programmers become reviewers of AI-produced code, the project’s speed will no longer depend on what the AI can write, but on what humans have time to review, and that is the step the data presents as the breaking point.
The March layoffs: Epic swore they were “not related to AI”
October’s pivot comes seven months after a round of layoffs. On 24 March 2026, in a note to employees published on Epic’s website, Tim Sweeney announced “over 1000” layoffs. The note attributed them to the downturn in Fortnite engagement that began in 2025, which means the company is spending “significantly more” than it makes, and added over $500 million of identified cost savings in contracting, marketing and closing some open roles.
Above all, the note said this: “Since it’s a thing now, I should note that the layoffs aren’t related to AI. To the extent it improves productivity, we want to have as many awesome developers developing great content and tech as we can.” On paper, October’s message is consistent with March’s: not fewer people, more output. What has changed between the two is not the announced headcount, it is the job. For the Disney game’s team, writing code becomes guiding the AI and reviewing what it writes.
In our investigation into layoffs “because of AI”, we showed that some of the jobs cut in the name of AI were already being filled again. Epic has taken the opposite path: keeping AI out of its layoffs, then putting it at the heart of its production. Both statements can be true at the same time. Only time will tell whether the “not fewer people” promise holds, whether or not the productivity gains materialise.
What video game developers think of it
Epic’s pivot runs into an industry that is increasingly hostile to generative AI, as the Game Developers Conference’s annual survey shows.
| State of the Game Industry 2026 survey | Result |
|---|---|
| Professionals surveyed | more than 2,300 |
| See generative AI as negative for the industry | 52% (30% a year earlier, 18% two years earlier) |
| See it as positive | 7% (13% a year earlier) |
| Negative view among programmers | 59% |
| Negative view among visual and technical artists | 64% |
| Already use generative AI in their work | 36% |
| Laid off in the past two years | 28% |
Source: Game Developers Conference, State of the Game Industry 2026.
The figures describe a contradiction: more than one professional in three uses these tools, and more than one in two thinks they are damaging the craft. The exchange of 18 October between Tim Sweeney and Kenney illustrates it. Kenney wrote: “Don’t let billionaires convince you slop is the future”, adding that they manage more than they make. Tim Sweeney replied: “This is bullshit, Kenney”, then: “I run a company. I write code. I contribute to published papers. I’ve upped my expectations of AI. Up yours!” Kenney answered: “Your employees are liking my post, they reach out to me all the time”, adding that they were sick of hearing “adapt or die”.
Tim Sweeney had already offered a reading of these criticisms on 12 October: “Much of the room wants it to be 2016 again, but it’s not.” For programmers, the question is not abstract: our analysis of jobs at risk from AI shows that, in the most exposed occupations, software development included, it is employment among beginners that is falling, not among experienced staff. Those are precisely the jobs where people used to learn the craft by writing code.
Unreal Engine: why Epic’s bet goes beyond Epic
Tim Sweeney said it in the very first post: the aim is also to bring the benefits “to all teams using Unreal”. And Unreal Engine is one of the most widely used game engines in the world, from big-budget productions to independent studios. What Epic learns on its own game will end up in everyone else’s tool.
That is already under way. According to Wccftech, Unreal Engine 5.8, the last planned major release of the current generation, will include an experimental plugin, disabled by default, that lets developers connect any large language model to the engine’s components: visual scripts, levels, materials, 3D meshes. Epic presents it as particularly important for Unreal Engine 6, where AI integration is set to become central to creation, and is targeting early access by the end of 2027, with a full release 12 to 18 months later. In the March note, Tim Sweeney already listed the move to Unreal Engine 6 among the company’s priorities.
The model named by Bloomberg is no accident either. Anthropic’s Claude is among the best-placed models for programming, as our coding agents ranking and our comparisons Claude Opus 5.5 vs GPT-6 Sol and Claude Sonnet 5.5 vs Muse Wonder Full show. But those rankings measure what a model can do on a test. They do not measure what a team of a hundred people manages to ship, review and maintain with it.
What to check over the coming months
Epic’s bet is now public, quantified and dated. It can therefore be judged on the evidence.
- The Disney game’s new date. Bloomberg says none has been communicated to the team. If the switch to AI delivers, the delay should shrink.
- The gains announced for October 2027. Tim Sweeney expects “very little” below three times January 2026 efficiency within a year. Will Epic publish measurements, and by what method?
- The final version of the launcher. It is the only vibe coded product Epic has shown. Its stability will be a first concrete indicator.
- METR’s new study design. The lab has announced it is changing the design of its study to correct its selection bias. Its next results will show whether the gap with self-reported gains is narrowing.
- Epic’s headcount. The “not fewer people” promise will be tested by the next rounds of layoffs, or by their absence.
AI already makes developers somewhat more productive, and probably more so than a year ago. But between the 18% time saving that METR measures at best and Tim Sweeney’s “5x”, there is a gap that nothing public fills today. And the only data available on the rest of the chain says that code produced faster is also more often broken, and less often reviewed.