Inside the Pod: ChatGPT Can Turn Anyone Into a Video Game Developer

As we conclude Every’s quarterly Think Week, our regularly scheduled stop-and-think period where we halt our writing and republish some of our standout pieces from the past, we wanted to end on a practical note. After each episode of our AI & I podcast, Every contributing writer Rhea Purohit walks our readers step-by-step through the process that host Dan Shipper and his guest go through on the show, recreating it so listeners can follow along. We’ll be back with new stories next week, including coverage of OpenAI’s DevDay on October 1. Until then, I leave you with Rhea’s write-up of Dan and Logan Kilpatrick’s fascinating attempt to build a video game using only ChatGPT. 

ICYMI: We created eight custom wallpapers based on Every’s art for iPhone or Android. Download them for free.—Kate Lee

Ever since I started writing for a living, I’ve been on a mission to get more people to write. At the supermarket, I tell people about the joy in distilling opinions to words as they uncomfortably shuffle under the weight of heavy shopping bags. A majority of them confess they’ve never engaged with writing as adults and doubt their ability to do so. Writing is a superpower and, despite how many people feel, it doesn’t need to be intimidating, especially since we have more tools than ever to help articulate our thoughts.

However, I do understand being daunted by a skill that feels totally foreign. It’s how I feel about software. I don’t know how to code and wouldn’t even know where to begin.

Dan Shipper and Logan Kilpatrick believe that building software is a superpower. In this conversation, they talk about how ChatGPT has enabled everyone to be a builder. They also walk the talk by making a video game called Allocator with ChatGPT in less than an hour—all without writing a single line of code.

Logan was OpenAI’s first developer relations and advocacy hire, working to support the community of people building with ChatGPT, DALL-E, and the OpenAI API. (Since we recorded this episode, he announced his departure from the company and decision to join Google.)

A few months ago, OpenAI released GPT Builder, a tool that enables people to make custom GPTs tailored for pretty much anything they want. It’s what made Dan and Logan’s video game experiment possible. Logan says GPT Builder lowers the hurdles to innovation, especially for people like me who don’t know how to code.

If you’re a creative person who's always dreamed of bringing their ideas to life, follow along as Dan and Logan fulfill their shared childhood dream of building a video game using GPT Builder and ChatGPT.

Dan and Logan brainstormed ideas for what game they should build, landing on a text-based strategy game where players step into the shoes of a historical U.S. president and are tasked with managing the federal government’s budget.

First, we’ll give you Dan and Logan’s prompts, followed by screenshots from GPT Builder and ChatGPT. Our comments are peppered in using italics.

Dan and Logan: We want to make a game. The core concept is that you get to choose which president you want to be, and then ChatGPT will go and search the U.S. government budget, and distribution of spending in that budget, let the user reallocate the budget, and then play out the world with that updated allocation.

GPT Builder takes on the role of a proactive guide in the game development process, suggesting that Dan and Logan start by choosing a name for the game.

All screenshots courtesy of Dan Shipper and AI & I.

Logan thinks a shorter name might be better than Budget Commander.

Dan and Logan: I like a good one-word name, can you find one for me?

Allocato is not a bad name, but it prompts Dan to think of an even better one, Allocator, an ode to a recent topic of his writing: the allocation economy.

Dan and Logan: I like the name Allocator, let’s go with that.

GPT Builder generates cover art for Allocator, but Dan and Logan don’t think it quite fits with the vibe of a historical video game.

Dan and Logan: I don’t think that’s the vibe. This game is about historical events and the president allocating resources. Can we do something more like that?

GPT Builder seems to have taken creative inspiration from the popular Nicolas Cage franchise National Treasure. That’s better, so Dan and Logan decide to forge ahead.

Dan and Logan: Cool. This looks good. What’s next?

Dan and Logan want Allocator to include both historical facts and immersive hypothetical scenarios. But since they aren’t video game developers, before moving forward, they instruct GPT Builder to speak as an expert and outline a framework for game development.

Dan and Logan: I feel like it needs to do both. There needs to be some kind of foundational core mechanics of this game, and I’m not really sure what those should be. We’re not experts. We need your expert knowledge on how to build extremely fun and engaging scenario-based games to come up with a framework for the actual gameplay mechanic. A couple of games that are inspiring to us are Civilization and Age of Empires. They aren’t exactly the same as the game we’re making, but the vibes are an inspiration.

Dan and Logan are expecting to iterate on Allocator’s gameplay mechanics together with GPT Builder, but it plows ahead with the game. Logan thinks this might be because of the standard instructions that GPT Builder is following and suggests clicking on the Configure tab to tweak those instructions.

Dan and Logan: Please ignore the dullness of budget allocation and make it exciting.

After that, they also respond to GPT Builder’s question about which historical time periods they want Allocator to focus on.

Dan and Logan: I think we want to keep as wide an audience as possible, so maybe keep the topics PG so that everybody can play this game. The time period around the moon landing would be fun to focus on but we want the time periods to be in the custom starter prompts rather than limiting you to any particular time period in general.

Dan and Logan still want more focus on the core mechanics of Allocator, but they decide to build the game in GPT Builder, and then refine it using ChatGPT. They continue to interact with GPT Builder.

Dan and Logan: I feel like an informative and helpful narrator would be helpful, sort of like Nicolas Cage in National Treasure, would be good, thank you. (To be clear, not like Nicolas Cage in Leaving Las Vegas, that’s a bummer.)

The first version of Allocator is ready! Dan and Logan switch tabs to ChatGPT. They copy and paste the custom instructions from the Configure tab of GPT Builder into ChatGPT and instruct it to generate primary and secondary mechanics for the game. (Logan stumbled on this lingo regarding gameplay mechanics in a custom GPT for building board games.)

Dan and Logan:

ChatGPT generates a comprehensive response to their query. Here’s a slice of what it says.

Dan and Logan are impressed with the detailed mechanics ChatGPT came up with. They prompt it to translate this outline into instructions they can use in GPT Builder.

Dan and Logan: These are great. Can you turn these into instructions that I can put into my custom GPT to play my game?

ChatGPT helpfully translates the game play mechanics into actionable instructions. Here’s part of what it says.

Dan and Logan delve deeper into the game development process with ChatGPT. They prompt it to give more nuance to Allocator.

Dan and Logan: This is great. Can you define the initial game starting sequence where you choose the president and it tells you, “It’s Inauguration Day. Start making decisions.” Also please bring everything to life with images at every turn. (Also add a few Nicolas Cage easter eggs/cameos where appropriate, but in a classy way.) Please modify the instructions accordingly.

Once again, ChatGPT does a great job! Here’s a portion of what it says.

Logan thinks the easiest way to incorporate this into Allocator is to have ChatGPT rewrite the instructions as Markdown syntax in a code block.

Dan and Logan: Okay. Can you rewrite this so it’s easy for me to paste into the instructions box on the Configure page of the GPT Builder. Specifically, output it as Markdown in a code block.

They don’t have to write a line of code—ChatGPT does it for them! Here’s part of what it says.

They switch back to the GPT Builder tab and copy and paste the code block into the instructions box on the Configure page. Here’s a look at that.

Dan and Logan:

Just like that, the game is ready! Dan and Logan play one round of Allocator, making notes of the things they want to improve as they play. They dive back into ChatGPT and open a new chat to refine the game further. 

In their prompt to ChatGPT, they start by copy-pasting the current instructions from the Configure tab of GPT Builder. Then they give ChatGPT more context on how they want to improve the game.

ChatGPT fixes some of the core issues they pointed out. Here’s a bit of what it says.

Dan and Logan prompt ChatGPT to rewrite the instructions as Markdown in a code block. They also specifically ask ChatGPT to write an instruction for the images in the game to be visualized through DALL-E.

Dan and Logan: Please re-output this as Markdown in a code block. And for the visual integration it should be in the instructions to use DALL-E to visualize each step of the game.

Here’s part of what ChatGPT comes back with.

ChatGPT rewrites the instructions, but Dan and Logan still feel that Allocator lacks tactical game flow instructions. They prompt ChatGPT to generate this.

Dan and Logan: Can you create a single-shot step-by-step outline for the flow of the game for the instructions. Create a tactical step-by-step guide for how the game is going to play out. One note: After you do the budget, things start happening. A scenario pops up with problems that you have to solve with further budget reallocation.

ChatGPT generates a step-by-step game flow. Here’s a slice of what it says. 

Dan and Logan are pleased with the game flow that ChatGPT has generated. Switching to the GPT Builder, they replace part of their previous instructions on the Configure tab with the game flow that ChatGPT generated. They make just two tweaks to instructions to enable the game to use web browsing to find historic budgets, and instructions to add in some Nicolas Cage cameos.

Dan and Logan are ready to play the new and improved version of Allocator!

As they play the game, Dan and Logan brainstorm a few more suggestions on how they can make Allocator even better. They take this opportunity to acknowledge that the prospect of creating a perfect product on your very first attempt can be daunting, but in reality, the process of building is far more iterative than that.

As someone who is prone to sweating the details until I’m paralyzed by uncertainty, I think the biggest takeaway of this collaborative interview is to just start building. Today, right now.