The 7% problem

The tools got bought and they sit mostly unused. You raised it in a meeting and the room went quiet in a way that was not about the tools. Then somebody forwarded you a survey, and the number in it was worse than you expected.

The survey is GDC's 2026 State of the Game Industry, more than 2,300 professionals. It found that 52% of them think generative AI is harming the industry. That figure is quoted everywhere, usually as the whole story. It's the least useful number in the report.

Start with 7%. That's the share who think generative AI is helping, and last year it was 13%. Run the negative line back and it goes 18%, then 30%, then 52% over three surveys. So the tools got better every year, and every year the people who make games thought worse of them. A better demo does not touch that. Nothing about the model does.

Break it out by discipline and it sharpens. Visual and technical art, 64% negative. Design and narrative, 63%. Programming, 59%. The resistance is hardest exactly where the tooling has come furthest, which is backwards from how adoption is supposed to go.

Then the number nobody writes about. Executives and business-ops respondents report 19% positive. The industry reports 7%. Usage tells the same story from the other side: 36% overall, but 30% at game studios against 58% at the publishing, support, and marketing firms around them. The people signing off on AI and the people who would live inside it may as well be at different companies.

And one of those companies has watched 28% of its people get laid off in two years, a third of them in the US. When someone in that seat hears that a tool is the future of the craft, they have already formed a view about who the future of the craft is for.

Handle these figures carefully, since we just published an essay about a statistic nobody checked. GDC customized the survey per group, so the sub-figures ride on different denominators and are not clean cuts of one population. The sample also dropped from over 3,000 to over 2,300, so read the trend as a direction, not a panel. And the line you'll see everywhere, "5% of studios use AI in player-facing features," is wrong. The 5% is a slice of use cases among people already using these tools, not a count of studios. Hold onto that one. Read correctly it's the most useful number here.

Adoption and adaptation are not the same problem, and a rollout that funds the first will skip the second. Adoption is whether people pick the tool up. Adaptation is whether the work around it changes shape: which roles shift, what the review step becomes, what "done" now means, how many people the job takes once the tool is in it. They are two different problems. The second is the one nobody budgets for, and it is the one that decides whether the first was worth doing. A studio can get every seat logged in and still be broken, because the tool changed the work and nothing else moved to meet it.

Read the artists' 64% through that and it stops looking like fear. If leadership wants the thing adopted but the roles hold, the deadlines hold, and the headcount holds, then the tool is not help. It is the same job with a faster way to be asked for more of it. The person who flagged it did the arithmetic and reported the answer.

This is why the usual playbook backfires here. The standard adoption motion is built for indifference: busy people, the old way works, the tool is one more login, so you fix it with training and defaults and a champion the floor respects. Point that at a team that has correctly read an adaptation problem and you have answered a question they did not ask. A session on prompting lands as management that did not listen. You spend trust and end up further from the outcome than when you started.

So look at what the people who kept using these tools actually use them for. Among users, 81% say research and brainstorming. Asset generation, 19%. Procedural, 10%. Player-facing, 5%. What survived contact with a studio sits upstream of the craft, in the thinking, not in the file an artist owns.

That is not a retreat. It matches where a live game makes its money. The work that pays sits in the analysis layer: which cohort is leaking, when the next event should land, what to test against churn and lifetime value. None of it touches anyone's art. All of it keeps the title alive, and it is the quietest ground in the building.

So we start there, and we do the adaptation work out loud. Before any tool goes in, we write down the boundary: which work is out of scope and stays out, and who is allowed to move that line later. Then the part rollouts skip. We say what changes around the tool, whose role shifts and how, what the new checkpoint is, what the team is now measured on, so nobody has to argue from a feeling six months on. The first deployment lands in the analysis layer, where the objection is thinnest and the payback is fastest, and it earns its way toward anything nearer the craft. Adoption gets scoped like a feature: named champions, training built into the rollout, usage targets you can see, runbooks so it holds after we leave.

This is not a consulting instinct. It comes from live games, where you cannot ship a system your players won't touch and call it a win, and where the surest way to lose a team is to tell them the thing they already know is wrong. Our founder ran products at PopCap, NaturalMotion and Wooga. Nothing in that history got a pipeline change adopted by arguing the people doing the work were being irrational.

If the tools are bought and sitting there, the read is probably not that your team is behind. You asked them to adopt something without adapting a single thing around it, and you asked as though the argument were already over. That is a sequencing mistake. It is fixable, and it is the first place we would look.