Shigeru Miyamoto was asked, at a press conference marking Japan's Order of Culture, what he thinks about generative AI in game development. His answer is the most useful thing I've read on the subject this year, because it separates two things the discourse usually merges.
He starts with an analogy. In the past, designers were taught to create beautiful lettering for posters, and that skill alone was enough to be called a designer. Or you could take a design that was trending abroad, give it a Japanese twist, and call yourself one. Then computers arrived, the lettering became a tool, and "a designer's true individual capabilities, such as their innate sense of layout, finally came to the fore."
Generative AI, he thinks, is a similar shift.
Creating and labor are different things
The distinction he draws is the core of it: "We distinguish between 'creating' something and the 'labor' involved in the process. When you set out to create, you first establish a broad structure, then begin the hands-on work: the tasks that follow, such as filling in gaps or refining the details."
That's a clean model. Creation is the structure — what the thing is for, how it works, what it refuses to do. Labor is the filling-in. AI is very good at filling in, and getting better at proposing structures, but the difference between the two is not a difference of difficulty; it's a difference of kind.
Then comes the warning, which is the best sentence in the interview: "The tricky part is that this labor can actually be quite enjoyable." With lower barriers to entry, some people will "mistake the satisfaction of building something for the achievement of creating something genuinely good."
That's exactly the failure mode of the current moment. The labor is now cheap and pleasant, and the pleasure of producing — the thing appears, it looks like a game, it runs — is easy to confuse with the judgment that the thing should exist in this form. The satisfaction is real. It's just not evidence.
Copy and paste doesn't last
His test is historical: "Looking back at past designs, there are some that were created merely by copying and pasting. Yet, such work rarely stands the test of time; it lacks the individual's particular design intention."
That's a checkable claim, not a mystical one. You can look at a piece of work and ask whether it carries a decision that someone made — a constraint that was chosen, a feature that was refused, a detail that exists because of a point of view. Copy-paste work, human or machine, tends to have no refusals in it. It has everything, which is another way of saying it has no intent.
The same test applies to AI-generated work. The question isn't whether a model made it. It's whether the thing has intentions in it that can be traced to a person, and whether those intentions are coherent enough to be legible.
What "there will always be a role" means
Miyamoto concludes that "even if AI were to eventually handle everything, there will always be a role left for us to play," because the process "relies on an individual's intent, with the computer serving as an assistant for the labor."
I think that's right about the value of intent and wrong to hear as a guarantee of employment. Those are different claims. Intent will always be the thing that makes work legible as someone's; that doesn't mean the number of people paid to supply it stays the same. The lettering designers of his analogy did not all become layout geniuses. Some of them lost work, and the skill that had defined them became a menu item.
That's the part of the analogy he doesn't dwell on, and it's the part the industry should. The optimistic reading — tools commoditize the low end, taste rises — is true in aggregate and brutal in particular.
The test I'd take from it
If you're building with these tools, the useful discipline is Miyamoto's split. Before the labor, write down the structure: what this is, who it's for, what it won't do. Then let the labor be cheap. And when something is finished, ask the copy-paste question — does this carry a decision, or does it carry everything?
The pleasure of making something appear is not the same as having made something worth appearing. That's true of every tool, and it's especially true of the ones that make appearing effortless.