During a recent sprint in the Play Lab, we were tasked with pushing the edges of the role that Generative AI can play in games. Armed with post-its, the latest AI tools, and IDEO’s homebrewed GPT prototyper, we designed and playtested a myriad of uniquely tech-enabled games, from board games where AI roasts your choices and quick scanning card games to immersive roleplaying mysteries with AI characters. In the process, we created a set of principles for better building with Generative AI ready.

1. Tighten iteration loops with Large Language Model prototyping

The first board game we designed involved an AI judge that roasts your in-game choices as you play. But on a fast-paced design sprint, we needed a way to quickly test and refine this idea to get to a working, and fun, game. The answer? Leveraging the instantaneous nature of prototyping with Large Language Models (LLMs) to help us get to “the one” quicker. 

We started the process with prompt engineering to quickly check our assumptions about whether the system could handle interpreting and responding with humor. By sharing that prompt with players, we were able to quickly test and iterate on a core interaction–submitting an action and getting roasted–from day one. 

Prompt engineering should be part of the design and prototyping process. Imagine testing interactive features with users from the outset and iterating in real time to unlock tight iteration loops. But remember, prototyping with AI is not a shortcut. Testing and iteration is about uncovering unexpected challenges as much as it is solving them. Which leads us to our next point:  It’s crucial not just to prototype with AI, but to define its role in the system.

2. Create more engaging systems by defining AI’s character and role

Don’t just consider the form and function of your AI—you  also need to think about  its role and character.  The more front and center AI is in a system, the more crucial it is to design for character and latency. For example, you’ll make very different design choices when you’re designing AI as a character that converses directly with you vs. designing for an AI as a foundational technology that assists in auto-completing your emails in the background.

So, start by asking yourself:

• Is the AI a facilitator, actively running (or even rewriting) the rules?

• Is the AI a player, interacting directly with humans?

• Is the AI a foundational tech, working unnoticed behind the scenes to enable specific features?

Once you define the AI’s role in the system, it is easier to map when and how it shows up, what kind of character it might have–or even if it should have one, if it is not in a foreground role— and how to create it.

Consider our AI Roast Bot. We needed the creative potential of Generative AI to come up with fun roasts, but that also meant less predictable responses. An unpredictable player can be fun, but a judge randomly and uncontrollably awarding or subtracting points less so.

To address this issue, we focused on creating the conditions for fun by mapping the AI’s and players’ roles to their level of control. Rather than having Roast Bot assume a game-controlling (but unpredictable) judge role, we let players strategically trigger the AI as part of their move. The unpredictability becomes an asset, as players have the agency to risk a roast—or score big. The AI Roast Bot’s ability to respond on subjective features (eg. creativity and wittiness) enabled us to craft a dynamic personality like Gordon Ramsay’s: scathing and hard to please. On the rare occasions that the AI is delighted, the payoff is extra satisfying. Thanks to AI, responses feel real and take into account players' unique choices to create a level of immersion and personalization previously only possible with a human judge. 

Iterating and defining the AI’s role in the system helped us design AI interactions that felt natural and enhanced the core experience, rather than feeling like a tacked on feature. But as fun as tinkering and iterating with AI can be, it’s crucial to keep the overall experience in mind, because AI is only one part of the whole. 

3. To keep people in the flow, design for the world around the AI 

No matter how front and center AI may be in your system, it is still only one element of the experience. 

When you’re bringing AI into the real world, suddenly there is more to think about than its response and tone. Whether you’re designing software or hardware, you might consider:

How do you interact with the AI? Does it actually help you accomplish your task? For example, text-based conversations with LLMs can give you rich outputs but not always in a form or manner that is convenient for your task.

• Does interacting with AI interrupt your task or switch your context?

• How does the AI get the data it needs? What are the input mechanisms?

• How long does the AI take to process and respond? How does that impact your flow?

When we were making a hybrid board game that mixes physical components with a digital AI Roast Bot, we encountered all these challenges—and more. Whether you’re making a game or a tool, it’s important to keep people in the flow, avoiding interruptions and distractions that take them away from the experience. By introducing the phone that housed our AI judge, we noticed it was easy to break the intimacy of the group and interrupt play rather than enhance it. To solve this problem, we zoomed out to consider the whole world of the experience and identify which elements are core to the experience, and which are distractions.

We started by leveraging quick ways to prototype and test the digital part of the experience, from prompt engineering to leveraging ProtoPie for a low-code, AI-enabled voice app. When we were playtesting with the text-based, prompt-engineered prototype, we noticed that players would crowd around the phone to read the AI's verdict. But for a multiplayer party game, that same interaction might end up excluding part of the group. 

When we considered the experience as a whole, we realized that if the AI is a player, it needs to be able to speak and interact like any other player to keep people in the flow. We switched to a voice-based interaction system, transforming the AI roasts  from one-to-one to AI-to-everyone. Not only did this make for more seamless input, we also discovered that it's way more fun to be roasted out loud by a silly computer voice than reading your own roast to friends!

Having AI interact with the real world on the fly is a challenge, but playtesting live prototypes and getting creative with latency issues helped improve the game overall. When you build AI-enabled experiences, the true value comes in designing for the big, unpredictable, and playful human world—and then identifying how AI can naturally fit in and enhance that experience. 

Come play with us!

Games are one of the best places to figure out how to creatively incorporate AI’s capabilities (and limitations) to enhance our experiences. The question we must ask ourselves is not how AI can make us more efficient, but rather, how we might leverage it to make our experiences more meaningful.

Imagine if we designed AI like games to create experiences that challenge us, guide us, and foster deeper connections. What role would you give AI? How might you design AI that supports human players to be the star of the game? The future of AI-enabled experiences depends not on the technology but the care and empathy with which we build them. Let's make that future tangible, playful, and unmistakably human.

Working on AI-related design challenges? Reach out to us at <helloplaylab@ideo.com>