Who’s Going to Supervise the AI?

The industry is worried about the wrong thing.

• 52% of game industry professionals think generative AI is bad for the industry, up from 18% two years ago

• Layoffs have hit junior roles hardest. Studios are having seniors absorb the work that used to develop junior talent

• The paradox of supervision: you need craft to supervise AI well, and AI is eliminating the work that builds craft

• Nobody is building a replacement pathway. Nobody is admitting that yet either

“Everyone is just having seniors do the work.”

A veteran game developer at Xbox said that to Wired earlier this year. It was offered as an observation about how studios are coping with layoffs and AI adoption simultaneously. It sounds like pragmatism. It sounds like studios making do. It isn’t. It’s a description of a pipeline eating itself.

---

What the industry thinks the problem is

The 2026 State of the Game Industry report, based on responses from more than 2,300 professionals, found that 52% of game industry workers now think generative AI is having a negative impact on the industry. Two years ago that figure was 18%. Only 7% think it’s having a positive impact, and that number has declined every year since tracking began.

The fears driving those numbers are real. Job displacement. Creative erosion. A quality floor that keeps dropping as AI-generated content floods storefronts. Studios using AI to justify headcount reductions they were planning anyway. These are legitimate grievances and they deserve serious attention.

But they’re pointing at the wrong problem.

The conversation the industry is having is about what AI takes away now. The conversation it isn’t having is about what AI takes away over the next ten years, quietly, from a generation of developers who never get the chance to become senior.

The paradox of supervision

Research into AI-assisted software development produced a finding that should make anyone who manages creative teams uncomfortable. Developers working with AI assistance scored 17% lower on code comprehension tests than those working without it. The more work they delegated to AI, the less capable they became of supervising the AI’s output. The researchers called it the paradox of supervision: the tool that amplifies your capability today degrades the capability you need to use it well tomorrow.

Game production has the same structural problem. Possibly worse. Code comprehension is at least testable. Production judgment isn’t. You can’t give a producer a comprehension test and find out whether they actually understand what a schedule is telling them, or whether they’re pattern-matching on surface features and calling it instinct.

Production judgment is built through accumulation. A specific kind of accumulation: doing things badly, recovering from them, and carrying the scar tissue forward.

What junior production work actually builds

I’m not going to romanticise the entry-level production experience. It’s mostly unglamorous. Chasing people for updates. Maintaining schedules nobody reads until something slips. Running triage meetings where you make calls on incomplete information and find out two weeks later whether you called it right. Preparing milestone builds and having a lead tell you that what you thought was done isn’t.

But that accumulation installs something. Dependency tracking you let slip and had to claw back teaches you what a schedule is actually measuring, which is risk, not time. Bug triage under pressure teaches you to make decisions without complete information and live with the consequences. Milestone preparation where someone pushes back on your definition of done teaches you the difference between activity and progress. Stakeholder communication when a project is in trouble teaches you to read what people aren’t saying.

None of that knowledge transfers through observation. You don’t get it by watching a senior producer do it. You don’t get it from a postmortem or a course or a framework document. You get it by doing it badly enough that it costs you something, and then doing it again.

That’s the pipeline. That’s always been the pipeline. Thirty years in production is thirty years of recoverable mistakes compounding into judgment.

“You don’t get it by watching a senior producer do it. You get it by doing it badly enough that it costs you something, and then doing it again.”

The pipeline is now broken from both ends

Here’s what’s happening structurally. Layoffs across the industry have disproportionately hit junior roles. 28% of the professionals surveyed in the GDC report were laid off in the past two years, rising to 33% for US-based workers. The roles that went first were the ones at the bottom: coordinators, associate producers, junior designers. Studios responded to the capacity gap by having seniors absorb the work. Which is where the Xbox developer’s observation comes in.

At the same time, AI is handling an increasing share of the work that used to sit at the junior level. Scheduling support. Documentation. Status reporting. Research and brainstorming. The GDC report found that the professionals most likely to use generative AI are business professionals and upper management, not rank-and-file developers. The tool is being adopted most enthusiastically by exactly the people who already have the judgment to supervise it, and applied most directly to the work that used to build that judgment in the people coming up behind them.

Junior roles are being eliminated from above by layoffs and from below by automation. What’s left is a generation of people entering the industry with fewer opportunities to do the painful, developmental work that produces senior producers.

The question nobody is asking

The industry’s anger about AI is focused on the present. Whose job does it take. What it does to creative work. Whether it degrades quality. All worth debating.

The question that isn’t being asked is ten years out. Who supervises the AI in 2036? Not in the abstract sense, but specifically: where are the producers with thirty years of recoverable mistakes behind them, the ones who can look at an AI-generated production plan and know, from the scar tissue, exactly where it’s going to fail?

That cohort is currently being trained, or it isn’t. The conditions that produce it are either in place or they’re not. And as far as I can tell, the industry is so focused on the immediate disruption that it hasn’t noticed the developmental pipeline quietly draining.

Nobody is building a replacement pathway. As far as I can tell, nobody is admitting that yet either.