In March, Electronic Arts laid people off at DICE, the Stockholm studio behind 'Battlefield 6'. The game had come out a few months before as the best-selling title in America in 2025, and sold seven million copies in its first three days. So the sequence reads: ship the biggest game the studio has made in years, then sack the people who made it. EA called this a "realignment", which is the word you reach for when "we are fine, this is fine" would not survive a follow-up question. The market did not punish it. Layoffs dressed as realignment rarely move a stock the wrong way.

Down the road sits Embark Studios, founded in 2018 by people who walked out of DICE to do it. Embark is the centre of the other big argument about AI in games this year. Both studios are a short ride from my desk. Not that it helped: I covered this from a chair, off the same reports as the press in Los Angeles. The chair was just in the right city.

The move is simple, and worth watching, because it will soon be run on a lot more people than game developers. A model cannot make a game. Not a good one, and not without a building full of humans steering every inch of it. Anyone who has watched a studio try to ship something with a chatbot at the wheel will tell you so for free. What a model can do is cheaper and quieter. It hands a chief executive a reason to take the humans out first and find out, afterwards, whether the work still gets done. The labour is not being replaced. The promise of replacing it is being used as the cover story for cutting the people, while whoever survives does the same job and trains the model that is meant to end them.


Take EA first, because EA is the one that says it into a microphone. Andrew Wilson, the chief executive, has called generative AI "the very core of our business" and floated that it could touch more than half of how the company makes things. Visionary stuff, on the slide. On the office floor it has meant a homegrown chatbot called ReefGPT, which staff are told to push their work through, and which by the account of employees who spoke to Business Insider returns broken code and confidently invented facts that a human then has to notice and unpick. The same people, on the art and design side, say they have been asked to train the model on their own work. So the job now includes teaching the software to do the job without you, on company time, with a smile. A former senior tester at Respawn, let go in the spring, thinks a model took over the part of his job that meant boiling down notes from hundreds of playtesters. He is probably right. Summarising is the one chore these tools do well.

EA reportedly told its developers that if you leave AI to its own devices, your work will not be exceptional. So the company ordering everyone to use the tool is also, on the record, admitting the tool alone makes mediocre work. It has decided to live there.


Down the road, Embark ran the other version of the experiment, the discreet one, where you ship the AI and wait to see if anyone notices. 'Arc Raiders' landed in October and was a monster: fourteen million copies by February, six million people in it every week. Buried under all that polish were the voices. The traders, the little robot barks when you ping a spot on the map, generated by text-to-speech and trained on recordings from actual actors who had been paid a licence fee for the privilege of seeding their own replacements. Players noticed. Eurogamer gave it two stars and aimed them squarely at the voices, which scored 'Arc Raiders' below 'Concord', the live-service game so unwanted its servers went dark a fortnight after launch. After enough noise, Embark's chief executive Patrick Söderlund went back, swapped a good chunk of the synthetic lines for human ones, and said the quiet bit out loud: "A real professional actor is better than AI. That's just how it is."

A screenshot of the 'Arc Raiders' trades menu, with the trader Celeste seated at left and a grid of purchasable crafting materials on the right.
Embark Studios, 'Arc Raiders' (2025). Celeste, founder and leader of Speranza and one of its five traders, at her trades screen. Several of the game's non-player voices were generated by text-to-speech trained on paid actors; Embark re-recorded some with human actors after launch. Image via SK Gaming.

The walk-back is the giveaway. Embark could afford to hire the humans back because on a game that sold fourteen million copies the voice actors were a rounding error. The text-to-speech was never a way to save money on one shooter. It was a rehearsal. Embark's owner, the Korean publisher Nexon, whose chief executive has cheerfully advised everyone to assume that every games company is now using AI, has waved 'Arc Raiders' around as proof that customers will pay regardless. Söderlund conceding that the humans sound better changes nothing about the plan. It just means the next outfit gets to run the synthetic voices on a game that cannot afford to fix them when the reviews come in.


None of this stays in Stockholm. Square Enix told investors in November that it wants generative AI doing seventy percent of its testing and debugging by the end of 2027, and announced layoffs across its Western offices the very same day, presumably to spare everyone the trouble of connecting the dots themselves. Playtika opened the year by cutting five hundred jobs, fifteen percent of the company, and said the plain thing for once: it wanted more AI and fewer salaries. Epic Games put a reported fifth of its own staff out in the spring. The Game Developers Conference runs a yearly survey, about as close to a head count as this industry keeps. The latest one found more than a quarter of respondents laid off inside two years, a third of them in the States, and over half of developers now convinced generative AI is bad for the work they do. Three-quarters of the students it polled are scared there will be no industry left to join. That is the figure that stops me. An entire trade is busy informing the people it has not yet hired that the lights may be off by the time they arrive.


The case for the tools is not stupid, and it deserves its best version before it gets taken apart. In a human hand, AI earns its keep. It grinds through the dull middle of an animation so an artist can spend the day on the shot people will remember, and it flags a crash at three in the morning that a tired person would walk straight past. Plenty of developers who loathe the hype use it exactly like that and ship sharper work for it. Todd Howard at Bethesda put the line where most people who make games put it: fine as a tool, not for generating the things themselves, because the human intention behind a decision is the part that makes the decision worth anything. Capcom has said it will keep AI-generated assets out of its games while still using AI to move quicker in the back office. Those are positions you can hold without embarrassment.

None of that, though, is what is going on at the companies shedding people. The respectable version of AI, the tool in the hand, has never once required firing the hand that holds it. EA did not let DICE go because a model can build the next 'Battlefield'. It cannot, and every person in that building knows it. EA swears the timing has nothing to do with the fifty-five-billion-dollar sale to Saudi Arabia's sovereign wealth fund, the one closing this year. Believe that if you want to. A company about to hand its new landlords the books has every reason to show them a smaller wage bill, and "we're leaning into AI" is the most flattering caption you can run under one. The tool is the alibi. The cut was always the plan.


The detail that stays with me is the artists being asked to feed their own work to the machine. There is something almost courteous in it. Nobody gets dragged out by security. You are invited, warmly, in the language of an all-hands, to give up a few hours a week to teach the software the thing you were hired to do, on the shared understanding that this is good for the company and, somehow, good for you. Maybe the model never gets there. Maybe it does. Either way the only lesson next quarter's numbers can teach is that the work got cheaper, and that the people who made it cheaper were in the room the whole time, being helpful. And none of this is a story about games. Games is just the part you can watch from the cheap seats, because the audience is huge and watching. The same memo is going round in newsrooms and call centres, in language every bit as warm. I keep thinking about how quietly a craft can be put down: across a long run of reasonable meetings, with decent coffee on the table and everyone in the room nodding along.