As AI models and tools improve, “taste” becomes the difference between slop and a marketable game. Some advice from some legends to help you develop it.
Ira Glass said something about creative work that’s been circulating online for over a decade, and it applies to AI game creation more directly than anything written specifically about games.
He said: all of us who do creative work get into it because we have good taste. But there is this gap. For the first couple of years you make stuff, it’s just not that good. Your taste is still killer. And your taste is why your work disappoints you.
That gap, between what you know is good and what you can currently produce, is what every creative person runs into when they start. But the AI game creation version of it is different from every other creative discipline that came before it.
The gap has never been smaller. And because of that, what matters most is not learning to code, not learning a specific tool, not learning game theory. It’s the taste you already have. The question is whether you know how to use it.
What Taste Actually Means in a Game Context
Taste is not the same as preference. Preference is “I like puzzle games.” Taste is “that puzzle game’s difficulty curve peaks too early, the player loses agency in the final third, and that’s why it feels unsatisfying even though the mechanics are technically sound.”
Taste is the ability to diagnose. Knowing not just that something doesn’t work, but why. Noticing the gap between what a game is and what it was trying to be, and being able to articulate it, even vaguely, even imperfectly, in a way that points toward the fix.
In traditional game development, taste was necessary but not sufficient. You also needed to know how to code, or model, or animate, or compose, or you needed the money to hire people who did. The technical barrier was high enough that plenty of people with genuine taste never made games because they couldn’t bridge the gap between knowing what they wanted and building it.
AI changes that. The technical barrier is now low enough that taste is the determining variable. The AI handles syntax while you handle feel. The AI generates the core loop while you decide whether the core loop is actually fun. The AI produces an enemy that moves and shoots correctly while you notice that the movement pattern isn’t surprising enough and players will tune out after the third encounter.
The 2026 Summer Engine review of AI game generators put it plainly:
“AI can build a working platformer. It cannot decide whether your platformer is fun. Pacing, feel, and difficulty curves still need a human in the loop.”
That human is you. The question is whether you’re doing that job, or accepting whatever the AI gives you and calling it done.
The Problem with Accepting Everything
The fastest way to make a mediocre AI-generated game is to accept everything the AI produces.
The output will be technically functional. Mechanics that work, assets that render, a loop that’s completable. What it won’t have is the thing that makes you want to play one more round. It won’t have the texture of decisions made by someone who cared deeply about how a player felt in a specific moment.
The GDC 2026 State of the Game Industry survey found 52% of professionals believe generative AI is having a negative impact on the field, up from 30% the year before and 18% the year before that. The people most opposed are those in visual art and game design: people who’ve spent careers developing exactly the kind of taste that AI can’t replicate. They’re not reacting to the tools. They’re reacting to the output of people using tools without taste, accepting whatever the AI produces, publishing it, and calling it a game.
That output has a name. Gameslop. Over 7,300 games on Steam carry an AI content disclosure label as of March 2026. Most of them exist because someone accepted what the AI gave them without asking whether it was actually good.
This is the context in which taste becomes decisive. The market is flooded with AI-generated games nobody edited. The ones that stand out will be the ones where someone cared enough to ask whether what they made was good, and then changed it until it was.
The Miyamoto Principle
Shigeru Miyamoto has been designing games for nearly fifty years. The through-line of his philosophy is simple: function first, form follows.
He doesn’t start with characters. He starts with how a player will feel in specific moments. The Goomba, Mario’s most basic enemy, exists because Miyamoto needed an enemy that beginners could defeat by stomping. He needed a specific function: a teachable defeat mechanic for new players. The visual, the name, the personality came after the function was clear.
Zelda started as “Adventure Mario,” a placeholder character in a binder of design specs for a game he was still figuring out. He worked on the feel first. The character who best suited that feel came later. As he told Time: “When I create a game, I try to focus more on the emotions that the player experiences during the gameplay.”
Justin Gary, designer of Ascension and Solforge Fusion and author of Think Like a Game Designer, names this same principle through the MDA framework: Mechanics lead to Dynamics, which produce Aesthetics. In his framing, Aesthetics are not visual style. They’re the emotional responses the game evokes. The framework is a reminder that mechanics are not the point, they’re the mechanism. The emotion is the point. Good taste, in Gary’s model, is the ability to trace the path from a mechanic to the feeling it produces, and ask whether that feeling is the one you intended.
For AI game creation, this means: know what you want the player to feel before you generate anything. Not the genre. Not what the main character looks like. The specific emotion you’re trying to produce in the player when they first encounter the core mechanic.
If you can answer that in one sentence, you have a standard to evaluate everything the AI produces. Does this feel like the thing I was trying to make? If yes, keep it. If no, change it. That’s how taste works in practice.
Five Ways to Develop It
Taste isn’t fixed. “You either have it or you don’t” is wrong. The fact that you can play a game and feel something is off, even without being able to explain it, means you have taste. What develops is the ability to use it more precisely.
1. Play games as an analyst, not a passenger
Most people play in passenger mode: experiencing what the designer created, reacting emotionally, moving on. Developing taste means switching to analyst mode some of the time. When you feel something, boredom, frustration, delight, the pull to play one more round, stop and ask why. What specifically produced that? What was the designer’s decision that created it? A sound effect? A timing? A camera angle? Something revealed or withheld?
You don’t need to analyze every moment. But you need to practice noticing when something lands and asking why. Over time that builds a vocabulary, a mental library of cause-and-effect relationships between design decisions and player feeling.
2. Make things you know are bad, intentionally
The fastest path through the Ira Glass gap is volume. Make a game in a day. Know it’s not good. Make another one. The purpose isn’t to produce good games. It’s to produce the experience of noticing what isn’t working, which is taste in action.
Gary’s core design loop runs on one principle: the faster you move through create, test, and revise, the better your game gets. Every iteration is a chance to notice a gap between what you intended and what the player experienced. That gap, identified and fixed, is how taste converts into skill. Gary applied this across formats, most recently with Solforge Fusion, a digital card game built on a decade of iterative thinking. The medium changes. The loop doesn’t.
AI game creation is the fastest version of that loop available to a new creator. A hypothesis about how something should feel can be tested in hours rather than months. That speed only matters if you’re making judgments at each iteration, not just generating and accepting, but generating, evaluating, and changing. The practice is the point.
A specific exercise: build a game with one deliberate flaw. Make the jump feel wrong. Make the enemy too fast. Make the level too short. Then fix just that one thing. The act of diagnosing and correcting a specific problem is the fundamental motion of taste. Do it enough times and it becomes instinct.
3. Write the emotional experience before you prompt
Before you open any AI tool, write one sentence describing what you want the player to feel in the first thirty seconds. Not what the game is about. What the player feels. “Curiosity with a slight edge of dread.” “The satisfaction of a system clicking into place.” “Momentum that makes you feel fast and in control.” “The warmth of something familiar being slightly subverted.”
Then build. When the AI produces something, measure it against that sentence. Not a genre convention. Not another game. The specific emotional experience you defined. This turns taste from a vague aesthetic preference into a functional design tool.
4. Steal like a designer, not like a player
There’s a difference between being inspired by a game and studying it. Players remember games they loved. Designers study games they admire, including ones they didn’t particularly enjoy, because they’re extracting mechanisms, not experiences.
Pick one game. Find what you think is its single best design decision. Articulate it precisely: not “the combat feels good” but “the parry window is generous enough to feel achievable but tight enough to feel satisfying, and the audio feedback is loud enough that success feels like a reward rather than an accident.” Then try to apply the underlying principle to something you’re making. Not the mechanic. The principle.
That’s how design vocabulary accumulates. Not by copying what other games look like, but by understanding why specific decisions produce specific feelings.
5. Show your work to one person and ask one question
Not “do you like it.” One specific question about the experience you were trying to create. “Did you want to try the next level, or were you satisfied stopping there?” “When the enemy appeared, did you feel surprised or did you see it coming?” “Was there a moment where you felt confused about what you were supposed to do?”
You’re not looking for general feedback. You’re looking for data on whether the specific feeling you intended is the feeling the player had. When there’s a gap, when they felt confused where you intended curiosity, or bored where you intended tension, that’s taste in operation. You know what you wanted. They told you what they got. Now you have something to fix.
Volume Is Still the Answer
There’s no shortcut to taste. Ira Glass knew this. Do a lot of work. Volume is how taste, the ability to see what isn’t working, converts into skill, the ability to fix it.
What AI changes is the economics of volume. It used to take months to make something worth evaluating. Now it takes hours. A professional team used to need a year to go through that cycle. Now it runs in a weekend. The judgment work is the same: notice what isn’t working, form a hypothesis about why, change it. But the volume of practice is now accessible in a way it wasn’t before.
The traditional developer’s version of taste was built through constraint. Limited memory forced decisions about what mattered. Technical barriers demanded clarity about what was essential. AI game creation imposes its own constraint: the tools are fast enough that the bottleneck is judgment. What do I change, and why, and how do I know when it’s right?
That judgment is taste. The way to develop it hasn’t changed. Make a lot of things, pay attention to what doesn’t work, and care enough to fix it.
Sources
- Ira Glass on Taste and the Gap — James Clear
- The Gap Between Having Good Taste and Doing Good Work — Kottke.org
- A Miyamoto Design Pillar: Forget Characters, Focus on Experiences — Game Developer
- Shigeru Miyamoto Quotes — TIME
- Game Creator Shigeru Miyamoto: The Philosophy and Ideas of Game Creation — Casa BRUTUS via Hayase (Dec 2025)
- AI in Game Development: Why the 2026 Backlash Is Real — TechBuzz AI
- 20 Best AI Game Generators and Tools (2026) — Summer Engine (May 2026)
- Vibe Coding AI Games: The New Way to Build Games by Feel, Not by Friction — Tesana AI (Jan 2026)
- How Vibe Coding is Revolutionizing Game Development in 2025 — Rosebud AI (Sep 2025)
- Think Like a Game Designer — Justin Gary (2018); also via Amazon
