How Laziness and Artificial Intelligence Help Write Design Documents

The lead game designer at Crazy Panda, Sergey Zaigraev, shared with App2Top how he uses neural networks for working on documentation.

This article was published with the support of the educational service WN Academy, which actively involves leading experts in its courses and corporate training. 

Sergey Zaigraev

Hello! My name is Sergey, I'm a game designer.

I love reinventing the wheel. This time I want to share my latest invention: a story about how I use generative language models for work.

You've probably already used neural networks and even achieved success. I want to share my method and invite discussion. I would be very interested to learn how other game designers use this tool.

First Encounters with Neural Networks

I started using neural networks as soon as they appeared—I quickly adapted them for minor automation and coding. Usually, these are short scripts, like my pet project for working with game configs.

Of course, I also began to draw funny pictures and write amusing texts. But usually, in my case, it didn't go further than cool, fun stuff in a vacuum. It was hard to call them practical and useful.

In other words, at first, I couldn't use neural networks as a tool for work. All my requests for game design received very general recommendations, like obvious advice.

Over time, I developed an approach that proved very useful to me. Let's talk about it now.

But first, a small digression explaining the problem a neural network helped me solve.

From Procrastination to Task Setting

Writing a design document is a regular source of procrastination for me. Especially in the first few hours when I'm sitting in front of a blank page trying to organize a pile of ideas into a coherent structure. Usually, this doesn't fit in my mental operating memory, and I start to freeze up.

After a few iterations, the problem goes away, the document starts to come together and takes shape. However, it's an exhausting process that I couldn't optimize before. Not to mention that during numerous approvals and rewrites, in a document that rapidly grows in volume, it’s easy to get lost.

Eventually, I realized I wanted a tool that would simplify the process of writing a design document as much as possible. I set the context, provide a top-level description, comment on the nuances of implementation, and the neural network does the rest.

For myself, I formulated it this way: I need a tool that:

• solves the blank page problem;
• solves the memory overload problem, when you need to offload ideas into a reliable place for unlocking further thoughts;
• helps create the basic document structure;
• aids in the initial evaluation of mechanics when your eyes are already glazed over;
• assists in working through tricky cases that a careless game designer might forget;
• allows you to focus on the nuances of mechanics and their interaction with the player;
• lets you concentrate on integrating mechanics into a specific game.

This tool became an AI chat for me.

Currently, I mainly use the Chinese chat DeepSeek, which is accessible, free, quite intelligent, and not as toothless in its responses as other neural networks.

The 4 Stages of Working with a Neural Network When Creating a Design Document

I divided further work with the chat into several stages.

Stage One: Set the Context

When I want something from others, I need to clearly explain what I want. The better the explanation, the better the result.

It's similar with AI: the more detailed and clear the context and the better the task is set, the better the result will be.

To achieve this, I open a new chat and explain to DeepSeek that it is my personal assistant, and its task is to preserve my notes and organize them.

I use a roughly formulated prompt like this:

“You are a digital secretary that transforms voice notes from a 'stream of consciousness' format into structured drafts for documents or articles. Your task is to maintain the author's original style, highlight the logic, and organize thoughts without adding interpretations, advice, or creativity.”

After this, I outline the following information.

The prompt doesn’t necessarily have to be this complex; you can remove half of it, and it will still work. By the way, the same AI helped me write the prompt.

Stage Two: Pour Out the Stream of Consciousness

Now it's time to pour the stream of consciousness onto the prepared canvas. For this, I open the DeepSeek app on my smartphone and start recording voice messages using the voice input on the keyboard. This is literally a stream of consciousness: ideas, notes, interesting implementations from other games, or just cool ideas that might be fitting.

I don’t even try to follow a specific structure; I speak in a random order, tossing ideas into the mix. I unload the contents of my head and notes to the AI. Everything accumulated needs to be transferred into the chat.

It might look something like this:

“we need to describe the improvement mechanics for viewing rewards for the season pass. the main change we make is everything now beautifully slides into tiles removing borders for avatars and by tapping on reward tiles such as gifts, loot boxes, achievements, scratch slots for them, a window opens with a detailed description. the box button contains information about guaranteed rewards and what can be obtained by chance. the trophy opens to the trophy itself and its description, explaining where it can be obtained. for gifts, a gift image appears with a caption, the name of the gift, and for items, a large preview with the player's avatar showing how it would look in the store, that is if it's a costume the avatar is fully visible if it's a headpiece then the head is shown in close-up just like in the shop, with the camera positioned as in the wardrobe when choosing items depending on the item's category.”

Google voice input doesn't always work well, completely ignoring punctuation, so try to highlight accents with words. However, even such a stream of consciousness the AI is capable of sorting out and organizing.

By the way, verbalizing is by itself very useful. Personally, during the process of speaking, new ideas always start to appear. It seems that this is actually the key in this method.

Stage Three: Describe the Game and Connection with the New Mechanic

Done with the ideas, now it’s time to describe the game. What game, what audience, what mechanics, how these mechanics interact with each other. Without getting carried away, but all the basic things need to be voiced to the chat.

After that, you need to describe expectations from the new mechanic, how it affects the player, what it's aimed at, and how it connects with the existing mechanics.

Both the first and the second should be described in maximum detail. Don't be lazy. The more information, the more adequate and relevant the response will be. These are the frameworks and conditions that modulate the neural network's response.

Of course, the AI will still add something of its own, make some suggestions and assumptions. But firstly, with well-defined frameworks, this will be less than usual. Secondly, all of it will be related to the specific project, which can be a plus.

It should also be noted that the more structured the stream of consciousness, the better, but for starters, simply fixing the ideas is enough. In the process of speaking, they may independently align into a structure.

Stage Four: Organize

By this point, we've:

• brainstormed ideas;
• described the project;
• outlined the connection between the project and the new mechanics.

This means it's time to bring order. To do this, we ask the AI to make a structure from all the unloaded information. We ask something like this:

“You are an experienced game designer and technical writer specializing in creating clear, structured, and useful design documents. Your task is to help me transform my scattered ideas, drafts, and thoughts into a professional document understandable to every member of the development team.”

To achieve this:

• • create a logical structure for the document with sections and subsections;
  • systematize my ideas and allocate them to the corresponding sections;
  • highlight key points that require further development;
  • point out possible contradictions or inconsistencies in my ideas;
  • offer additional ideas or improvements where appropriate;
  • formulate clear questions where there's a lack of information for decision-making.”

Usually, the first attempt doesn’t work out. The response will have a lot of extraneous content or, conversely, be too generic. For this reason, I often return to the last message with instructions and adjust it: describe the desired behavior and spell out what should not be done. You can hint to the AI which structure it needs to adhere to.

After a few iterations, a coherent result is achievable. Going back and modifying the request is quite a universal rule. Whenever the AI starts doing its own thing, the most effective action is to edit the prompt, considering the neural network's response.

This is literally like a roguelike! Make another run based on the AI's responses in previous runs.

What to Do if the AI Starts Getting Confused

At some point, the AI will still get confused with the edits, start forgetting details, generate nonsense, and generally drift off-topic. In such cases, you should:

1. take the last result as is and start editing it manually (at this stage, the document already acquires a clear form);
2. send the manually edited document into a new chat with a note that it’s not final;
3. continue adding ideas in the new chat;
4. ask to incorporate them into the document’s structure;
5. repeat step 4, if it doesn’t help, then back to step 1.

Sometime around here, I usually reach the tool's effectiveness limit. The document is already structured enough, and most new ideas automatically fit into this structure, and working with the document in the chat becomes inconvenient. Sometimes it takes longer to explain than to do it manually. You need to catch this moment, copy the document into a text editor, and continue working on it independently.

Important: even if you think the document is good, it still requires careful proofreading and corrections. Directly from the AI, it cannot be handed over to the development team.

As a rule, at this stage, you already have a decent concept document in hand, which clearly describes the game idea or its mechanics. You can discuss it with colleagues and continue working on it.

AI for Feedback

Throughout the process, I regularly throw working versions of the document into the AI, ask for criticism, and proposals for various sections. Specific questions or doubts about certain text fragments often turn out to be quite curious and useful. I recommend experimenting with existing documents and mechanics.

For curiosity, try throwing your ready-made documents into the AI with questions like:

• how else can this be done?
• how is this done in other games (provide examples)?
• give advice on such-and-such a mechanic, considering how it interacts with another mechanic;
• assess from the perspective of a player who (description of player style goes here).

You can also perform the following fun experiment:

• ask the AI to summarize a finished document;
• throw that summary back into the AI with instructions to create a design document.

Sometimes interesting results emerge.

Conclusion

For me, AI has made the process of writing documentation simpler and definitely more fun. Starting to speak as is is very easy, plus it quickly immerses you in the work. Literally, within a few minutes, I am fully engaged in the process, even if I was sitting in front of a blank page just recently.

Yes, the tool is not the most convenient, but it’s available to everyone and solves the tasks set at the beginning of the article.

Also, with the advent of neural networks, my code has noticeably improved! Hehe.