The week the games industry got exactly the AI it paid for

Gaming 3 min 80 sources

Epic and big publishers leaned hard into generative AI to cut content costs — and the bill came back as player distrust, fixing-up work, and partners walking away.

Key takeaways

Epic and big publishers pushed generative AI hard this week to cut the soaring cost of making games.
The behaviour it bought wasn't what they intended: AI mistakes its own artists must fix, players who reject it, and partners like Vampire Survivors' studio reviewing whether to walk away.
Layoffs the same week — including a praised nine-person team cut a month after launch — show the cost pressure that makes the gamble tempting, and a Life is Strange studio is weeks from running out of cash.

This week the games business made its clearest bet yet on generative AI — software that produces art and assets from a prompt instead of a person — and got a fast, awkward answer back.

Epic puts AI at the center of its engine

At its State of Unreal 2026 showcase, Epic Games said generative AI would play a much bigger role in Unreal Engine 6, the toolkit a large share of the industry builds on [3][7]. Epic then published a video showing how it already uses these tools inside Fortnite: an artist draws a character by hand, then an AI prompt reshapes it toward a finished 3D model [16][23].

The same video showed the catch. The AI riddles Fortnite characters and buildings with mistakes that Epic’s own artists then have to fix [23]. Epic laid off about 1,000 people roughly three months ago [23]. The promise of the tool is fewer hours per asset. What it visibly produced this week was more rework on top of fewer workers.

The people who’d have to use it said no

The reaction landed quickly, and from inside the industry rather than the comment sections.

Poncle, the studio behind the indie hit Vampire Survivors, said it was “reviewing” a just-announced Fortnite crossover after Epic’s AI reveal — a lucrative deal it now seems willing to walk away from [17][32]. (“Crossover” here means putting one game’s characters into another, usually a paying partnership.) The communications lead at Pocketpair, maker of Palworld, put it flatly: “Gamers don’t want it,” and said the studio doesn’t touch generative AI [5].

The sharpest warning came from someone who ran AI at a major publisher. Take-Two’s former head of AI said the generative-AI hype is “poisoning the well” — pushing it so hard that players may come to reject all AI in games, even the older, uncontroversial kind that’s quietly run matchmaking and physics for years [65]. The worry isn’t that the tool fails. It’s that chasing it burns the trust the business runs on.

Why the industry reached for it anyway

The pull is real, and it isn’t mysterious. Epic CEO Tim Sweeney said it himself: “a tidal wave is sweeping over the AAA game business” [42]. (“AAA” means the biggest, most expensive games.) Budgets have ballooned, and every studio is under pressure to make more for less.

So the industry rewards what it can measure — cost per asset, speed to ship, content volume — and generative AI scores well on exactly those. The Eidos Montreal founder named the shift bluntly this week: modern executives are driven more by spreadsheets than by passion for the games [50]. A tool that makes the spreadsheet look better gets adopted, whether or not it makes the game better.

The same week’s layoffs make the squeeze concrete

The cost pressure isn’t abstract. The entire nine-person team behind Luna Abyss, a shooter that reviewed well (81 on Metacritic), was laid off a month after launch [6]. They’d spent seven years on it; the game peaked at 317 players on Steam [6]. EA is reportedly cutting recruitment, customer support, safety, and IT staff [10][12]. Good work and good reviews didn’t protect either team. The math that closes studios is the same math that makes an asset-generating prompt attractive.

The under-covered one: a studio counting weeks of cash

Away from the AI fight, Life is Strange studio Don’t Nod is in genuine trouble. Its own auditors warned it will run out of cash by November unless it secures new financing [13][34]. It’s a reminder that the pressure driving the whole AI gamble is survival — and that for some studios, the clock is already this short.

02 · Lesson · why it matters

You get the behaviour you pay for, not the one you wanted

Reward a number and people will move it — even when moving it works against the thing the number was supposed to stand for.

A tool that scores well on the wrong test

Epic showed a video this week of generative AI building Fortnite characters. Watch closely and you see the AI making mistakes that human artists then have to fix. Epic ran that as a feature, not a confession. To understand why, stop looking at the tool and look at what it’s being graded on.

A modern studio is graded on a handful of numbers: cost per asset, time to ship, how much content goes out the door. Generative AI scores beautifully on all three. It is cheap, it is fast, it produces a lot. If those are the numbers that decide whether your team survives the year, the tool isn’t a temptation. It’s the obvious move.

The goal and the measure are not the same thing

Here is the quiet trap. Nobody at a game company wants “low cost per asset” as their actual goal. They want a game people love, a studio that lasts, players who come back. Cost per asset is just a stand-in — a number that’s easy to track and roughly points at the real thing.

But you can’t reward a goal you can’t measure. You can only reward the measure. So the measure is what people optimize. And the moment a tool moves the measure hard in the right direction while quietly pulling the real goal the other way, the system can’t tell. The spreadsheet sees a win. The Eidos Montreal founder named it plainly: executives are now driven more by spreadsheets than by passion for the games. He wasn’t describing villains. He was describing people doing exactly what they’re rewarded to do.

The behaviour comes with everything attached to it

When you pay for output, you get output — and you also get every side effect that rides along with it, whether you wanted it or not.

Epic wanted cheaper art. It got cheaper art, plus AI errors its remaining artists now spend time correcting, after the company laid off about a thousand people. It wanted a thriving engine everyone builds on. Instead the Vampire Survivors studio is “reviewing” a lucrative Fortnite crossover, and the Palworld team says simply, “Gamers don’t want it.” Take-Two’s former AI chief warned the push is “poisoning the well” — souring players on AI in games entirely. None of that was the intended behaviour. All of it was bought with the same decision.

The people who set the reward are rarely the ones who feel it

There’s a second layer. The person choosing to reward cost-per-asset is usually not the person whose work gets reshaped by it, and not the player who notices the seams. An executive reads the quarterly number. An artist absorbs the rework. A studio like Luna Abyss — nine people, seven years, a game that reviewed at 81 — gets cut a month after launch because the number that mattered, players on day one, came in at 317.

The reward and the consequence land on different desks. That’s why a system can keep optimizing a measure long after it’s stopped serving the goal: the feedback that something’s wrong arrives somewhere the decision-maker isn’t sitting.

What this looks like from the outside

You don’t need to run a studio to be inside this. Anywhere a number stands in for something that matters — a school graded on test scores, a hospital on wait times, a worker on tickets closed, a writer on clicks — the same shape repeats. Reward the proxy and people will deliver the proxy, brilliantly, including the parts you’d have begged them not to.

The honest move isn’t to stop measuring; you can’t run anything blind. It’s to hold the number a little more loosely — to keep asking what behaviour it’s actually buying, and to remember that the win on the page and the thing you wanted are two different objects that only sometimes point the same way. This week the games industry got exactly the AI it paid for. Whether that’s the AI it wanted is a question the spreadsheet was never built to answer.

03 · Lab · your turn

What You Pay For

Pick the number a studio rewards and watch the behaviour it actually buys, side effects and all.

04 · Hope · carry this

The encouraging part of this week is who said no. When a system rewarded the wrong thing, the developers and players closest to the work were the ones who noticed — and that instinct, to care about the game more than the number, is the very thing no spreadsheet can buy and no quarter can erase.