AI Product Design Through the Lens of Games

SANTORI
LABS
the humans
behind tempo
8 Apr '25 · 15 min read
AI Product Design Through the Lens of Games
Why LLMs' Probabilistic Nature Requires New Mental Models
Serai
Yuting
Image generated using Midjourney
This article is part of our series on the evolution of human-computer interaction in the era of Generative AI.

The existing product design playbook is breaking.
Design Thinking, Double Diamond, Lean Product Development - these methodologies that we lived and breathed are failing us in the age of AI. The evidence is everywhere: AI products today are built by slapping a chat interface on existing products and following the old playbook. No wonder they just feel like slightly smarter versions of what came before.
Our past year was also spent applying all our usual methods - personas, problem statements, existing workflows - they don't work well anymore. The question we keep asking ourselves is: are we trying to build a car, or just faster horses?

Henry Ford's famous quote about innovation and paradigm shifts.

Let's time travel to 2021 - how would we design our way to ChatGPT?
Here's the fun part: ChatGPT wasn't 'designed' in the traditional sense. There was no designer + product manager + engineers + researchers. It was created or stumbled upon as a product by releasing a simple yet intuitive chat UI over GPT-3.5.
If we had followed conventional product playbook:
  1. Define personas: Who are the users we should target? - Everyone!
    • (When your TAM is literally "anyone who uses language" and try explaining that to your product manager)
  2. What problems do they have? - Multi-turn conversations? Writing docs? Code?
    • (Try writing a spec for "have a conversation about anything")
  3. What is their existing flow?
    • (That's like asking someone in 1880 "how do you make sure you have light 24/7?" and they'll describe their system for maintaining kerosene lamps, storing oil, trimming wicks, and keeping spare candles...)
Innovation often requires breaking down the conventional processes - after all, you don't build a car by iterating on horse carriages.
This is especially true with AI, where we're dealing with something fundamentally different. At their core, LLMs are probability engines, which means every interaction has countless possible paths. Unlike traditional software where we can map out exact user flows, AI interactions are inherently probabilistic.
This makes traditional product development - with its emphasis on defined user journeys and predictable outcomes - fundamentally insufficient. We need a different way of thinking.
This is where game design comes in.
I've lived in both worlds - first as an game creator building experiences from scratch, then as an engineer building consumer software products. Creating games taught me how to design for emergence and open-ended interaction, while building products showed me how these principles could extend beyond games.
The intersection is particularly relevant now: games have always dealt with probability and player freedom - exactly what we need for AI products.

Game Design Principles for AI Products

1. Operant Conditioning

Ever wonder why you can't stop playing Tetris? Or why clearing your inbox feels so satisfying?
It all comes down to a principle psychologists call operant conditioning - where behaviors are shaped by their consequences. The classic example is the Skinner Box, where a rat learns to press a lever for food. Games are really just elaborate Skinner Boxes, but instead of food pellets, they reward us with something even more powerful: dopamine hits.

The Skinner Box experiment demonstrated how behaviors can be shaped through reinforcement.

The rewards in games are defined nebulously. An obvious interpretation is giving players loot, collectibles, or achievements. But that alone is a shallow understanding of the concept. A reward is really just anything that gives the player a Dopamine hit.
How do we do that? Well, if the player does something you want them to keep doing, such as clearing a row of blocks in Tetris, you can:
  • Play a satisfying sound effect
  • Show a particle-filled visual flourish
  • Give players more points
  • Allow further progression into the game
Each of these results in a Dopamine hit. Do them all at the same time? The player's behavior will be strongly reinforced and they will want to do that all over again.
The magic of great design isn't forcing users down a path - it's making the right path feel like a revelation.
While games are inherently structured around rewarding desired player behaviors, product design can also leverage this principle to encourage certain actions and discourage others. For user flows or interactions you want to promote, the UX should feel smooth and gratifying. For example, make a button that triggers a key functionality large and visually appealing, and accompany the click with delightful visual effects and sounds. In contrast, actions you wish to discourage could be obscured, and do not provide satisfying feedback when performed.
This isn't just about games. Think about your favorite apps - the ones you keep coming back to. They're probably not the most feature-rich, but the ones that make every interaction feel satisfying. A perfectly timed animation, a subtle sound effect, the way elements respond to your touch - these aren't just decorative. They're carefully crafted rewards that make the right actions feel natural and rewarding.

2. Onboarding

Players hate reading.
Although game designers all want their players to play the game correctly, starting them off with a lengthy dissertation on how to play will often lead players to just balk and give up. Yes, some people enjoy reading lengthy rulebooks, but most players really don't.
Game designers can't just tell people how to play - they have to show them.
So instead, games teach through the problem-tool-solution paradigm. First, players are presented with a problem, like needing to cross a chasm that they can't just walk over. Then they're given a tool, like a jump button. Finally, the player will need to apply that tool in order to solve the problem, which in this case is correctly timing the pressing of the jump button which will let them cross the chasm. Note that we never told the player to jump across the chasm; they are simply denied progression (and Dopamine) until they figure out how to use the jump button to cross the chasm themselves, thereby internalizing the solution while also having fun in the process.

Super Mario Bros. demonstrates the problem-tool-solution paradigm: gaps between platforms (problem) are solved using the jump button (tool), teaching players through direct experience rather than instruction.

This principle can affect how we think about product onboarding. Instead of overwhelming users with a complete UI tour, we can introduce core interactions one at a time. For example, first we let users write a paragraph in a document, and only after they've gotten comfortable with that, we introduce the AI proofreading feature. Each new tool builds on what they already know, making complex features feel like natural discoveries rather than overwhelming options.

3. Mastery

Good games ensure a single tool like the jump button can be used to solve multiple problems, like reaching new areas or defeating enemies by jumping on them. Using a familiar tool in a new context creates player enjoyment and reinforces their mastery of the tool.
Games may also layer several mechanics that can be combined in different ways, like jumping while running, or jumping off of walls to wall jump, or jumping onto walls and then running along the walls. This allows for combinatorics to kick in, where players can combine a very small set of tools to create a vast pool of possibilities that solves many different problems. Let's look at chess, a game where how the pieces move are simple universal rules, but in spite of that simplicity, players take years to develop not only mastery, but also self expression through their own distinct play styles. And as a humbling reminder, players will combine the tools in surprising and unexpected ways, discovering solutions the designers will never be able anticipate.

Chess exemplifies mastery through simple rules: each piece's movement is straightforward, yet players develop complex strategies and personal playing styles over years of practice.

To map this back to product design, we can ensure that the tools we provide our users are not bespoke one-off solutions that address just a single problem. Instead we can build tools that leverage combinatorics to work with each other, and believe in the users' ability to discover their own workflows. Humans are optimization creatures, so if you give them the tools and the space to explore, they'll figure out what uniquely works best for them, even without you building explicitly for that user journey.

Caveat - Games Are Not Apps

Here is something that is important to remember: Games and apps are different.
An app is a tool meant to solve the user's problems. For example, if you need to write an essay, you open a writing app and get to work.
A game, on the other hand, is different. The designer creates both the problems and the tools to solve them. And somehow, solving these artificial problems becomes fun.
Because of this difference, we cannot simply build a product like it's a game. Yet many look at games' engaging power with envy, and try to replicate it in their products. The most obvious attempt? Gamification.

Gamification in apps: Play store points and badges.

hot take
The Gamification Trap
What is gamification? For example, in a writing app, if the user wrote 10 documents with 1000 words or more using AI, they get a bunch of experience points and a "super AI user" rank as reward.
However, this is an extremely lazy way of using the principles of game design. It's just abusing the user with Skinner Boxes. This hijacks the user's intent to serve the designer's own desires, creating a hostile relationship between the user and the app. Think about just how many people have said if they weren't so hooked, they'd quit social media if they could? That's a hostile relationship. And usually this is done to drive engagement metrics and, hence, monetization. This undermines the fundamental idea of a tool existing in order to solve the user's problem. So stop. Don't do this.
The real opportunity in AI products isn't about gamification - it's about helping users learn, master and navigate complex AI-enabled tools and systems. As we explored in our previous post, the goal is empowerment, not engagement metrics.

4. The Core Game Loop

A huge issue that games deal with is combinatorial explosions.
What's that? Well, games are all about giving choices to the player. But if a player is simply given 5 possible choices to make 3 times, there are suddenly 5^3 = 125 possible journeys they could've made. That's a combinatorial explosion - too many possible journeys for designers to map out linearly.
To make sense of all this infinity, designers use what we call a core game loop. We start with the player in a default state and map out the decision tree of all the possibilities they could explore. But eventually we want to bring them back to that default state on all possible paths. This way we can model the player's behavior as a loop, where from the default state the player can try path 1 through 5, but they'll eventually come back to start the loop all over again. Now designers can work with the finite instead of the infinite.

Elden Ring's core game loop demonstrates how complex games can be broken down into simple cycles: from an idle state, players either explore for loot or engage in combat, always returning to the same starting point regardless of success or failure.

This maps perfectly back to what we discussed about LLMs earlier - as probability engines, they generate each token by predicting what comes next. Traditional product design tries to map every user journey - but with AI, that's like trying to predict every possible conversation. Game loops give us a better way: instead of mapping infinite paths, we create natural cycles that give structure to emergence.
The key isn't controlling every possibility - it's creating a framework that makes exploration feel natural.

Creating Spaces, Not Paths

These principles from game design - operant conditioning, guided discovery, mastery through combinatorics, and core loops - offer useful ways to think about AI products. Not because we want to make everything into a game, but because games have always dealt with emergence and probability.
The challenge isn't just that our old tools don't work - it's that our old ways of thinking don't fit. We can't map every path through an AI interaction, just as we can't script every possible chess game. But we can create systems that make exploration natural and mastery possible. That's what game designers have always done.
Perhaps that's the real insight: the best games don't try to control the player's every move - they create spaces where discovery feels inevitable. As we build this new generation of AI products, we're not just designing interfaces. We're creating environments where humans can explore, learn, and find their own path through the possible.
we're wondering...
What happens when we shift our mindset to defining the environment and providing tools to users, and trust them to explore and discover the use cases? What if that is a part of the new design thinking processes?
Reach out if this resonates - we're always looking to learn from others building in this space.

References

  1. Foundational Works
    • Skinner, B.F. (1938). The Behavior of Organisms: An Experimental Analysis. - The foundational work on operant conditioning that influenced game design.
    • Engelbart, D.C. (1962). Augmenting Human Intellect: A Conceptual Framework. - Pioneering work on human-computer interaction and augmentation.
  2. Game Design Theory
    • Salen, K., & Zimmerman, E. (2004). Rules of Play: Game Design Fundamentals. - Comprehensive framework for understanding core game loops and player engagement.
    • Crawford, C. (1984). The Art of Computer Game Design. - Early insights into game design principles and player psychology.
    • Schell, J. (2008). The Art of Game Design: A Book of Lenses. - Modern perspective on game design principles and player psychology.
  3. Design Methodologies
    • Design Council. (2019). The Double Diamond: A universally accepted depiction of the design process. - On the double diamond design process.
    • Cooper, A. (2004). The Inmates Are Running the Asylum. - On personas and user-centered design.
    • Blank, S. (2013). The Four Steps to the Epiphany. - On lean product development and customer discovery.
  4. Related Reading
← prev
The Augmented Mind: Engelbart's HLAMT Framework in the AI Era
Stay in touch
Thoughts on AI, knowledge work, and the future of human-computer interaction.