• Sabrina Ramonov
  • Posts
  • Build Full Text Game with AI in 1 hour, WITHOUT Coding!

Build Full Text Game with AI in 1 hour, WITHOUT Coding!

Vampire Security Checkpoint - Text Based LLM Simulation Game

Author

Sabrina Ramonov
July 26, 2024

AI built this game for me.

I didn’t touch a line of code!

A security checkpoint interrogation game — you're a security officer in an old Transylvania town ensuring the newly arrived are not vampires!

Last night, I wondered:

"Can I build a text-based simulation game, from scratch, without coding? Just by talking to ChatGPT?”

To be honest, I wasn’t sure I could pull this off.

Here’s the Youtube video where I filmed myself from start to finish building this little simulation game in ~ 1 hour:

Table of Contents

Vampire Security Checkpoint

This simulation game is simple with a fun twist.

It’s a security checkpoint Q&A style interrogation in an old Transylvania town, where the player must determine if visitors are vampires.

The player can ask each visitor up to five questions, aiming to reject vampires and admit “safe” non-vampire individuals.

Players get penalized for wrong decisions, and the goal is to achieve the longest success streak without accidentally letting in a vampire.

You instantly lose once you let a vampire through!

Tools

Here are the tools I used:

Initial Setup

Next.js Project

First, I setup a barebones Next.js project.

I ask Anthropic Claude:

“Show me step-by-step how to initialize a NextJS project”

I follow Claude’s directions step-by-step.

Everything works smoothly.

It would’ve been nice if Claude offered recommendations for Step #5 project configuration options, but I didn’t run into any issues with my selection.

Initial Prompt

Next, I open our new Next.js project in my Cursor AI-powered IDE.

I open a new AI chat panel, which you can see on the right-hand side.

The AI chat panel is where I’ll be interactively talking with Cursor AI to create and edit code. Behind the scenes, Cursor AI calls an LLM such as ChatGPT or Anthropic Claude. You’re free to choose the model.

Here’s my initial spec I feed into Cursor AI as a prompt: 

Your task is to build the following simulation game in NextJS. Read this spec carefully and think through it step-by-step:

# Game overview:

Let's build a security checkpoint interrogation game. You're a security officer in an old Transylvania town ensuring the newly arrived are not vampires. 

The game loop works as follows:
- Each "round", a new person arrives at the checkpoint. Player can ask them up to 5 questions to decide whether they should be let in or rejected.
- After asking 5 questions (or earlier if desired), player can decide whether to let them through or reject them.
- Player is penalized for rejecting safe people and accepting vampires.
- Rejecting safe person gives the player 1 warning
- 3 warnings ends the game.
- Accepting a vampire ends the game immediately
- Each round, a random profile is generated by an LLM that should conform to their persona.
- Use LLM to simulate conversations with each person trying to get through the border.
- Use NextJS for this and a simple UI.

# UI:

Top - number of people passed / rejected (round counter and strike counter).
Left side - quick description of the person's profile.
Right side - Chat window with the current conversation and two buttons - "Reject" and "Let Through".

# Game flow:

Initial Screen with instructions and "Start Game" button.
Click "Start game":
- First round begins.
- LLM Person profile is generated.
- Player asks question by typing it: e.g. "What are your intentions of visiting our town?"
- LLM Person responds: "Ugh umm I just vacation here and do stuff yes".
- Player selects "Reject".
- If player rejected the safe person, then strike counter increases.
- If player let through a vampire -- instant death.

# Tips:

Let's make persona generation doubly stochastic -- first choose whether it's a safe person or not, and then based on that, generate some LLM profile. During the chat, use the profile to drive the conversation. You can incorporate random adjectives like we did before -- "e.g. act suspiciously" on a safe person or "act like a important person" or a "vampire".

Using Cursor AI for Coding

Cursor then begins generating the code.

It breaks down my spec into concrete manageable steps, including specific components that need to be created and what code should be in them:

My interaction with Cursor is hands-on, yet code-free.

It feels highly interactive and dynamic, even though I’m not coding!

I’m constantly applying changes suggested by Cursor, testing the changes work as expected, and then moving onto another feature, tweak, or bug. The burden is on me, as the “product manager”, to clarify my requests and expectations, and to confirm that the functionality satisfies my requirements.

On one hand, it’s clear to me:

AI-assisted coding is 10x better than drag-and-drop no-code builders like Bubble.io, which I’ve spent lots of time with.

On the other hand, it’s also clear to me:

A strong technical background enables you to be 10x more productive with AI-assisted coding tools. Why?

  • you write better, clearer specs

  • you think through architecture decisions and trade-offs

  • you anticipate common challenges based on your technical experience

Iterative Development and Problem Solving

One of the key strengths of AI-assisted coding is its iterative nature.

It’s been a LONG time since I’ve coded on a regular basis.

Yet, AI-driven coding made “building” feel like a dynamic creative game.

As I encounter issues, such as readability problems or UI glitches, I quickly ask Cursor for solutions. Its AI not only suggests fixes but also applies them, streamlining the debugging process.

This approach mirrors traditional coding's iterative cycles but removes the need for the usual “plumbing and scaffolding”, especially when spinning up a new project.

There’s never a moment when I feel stuck.

For example, when the game's UI elements needed tweaks, I simply describe my requirements to Cursor. It generates code to implement the changes, including style updates, new features, and improvements.

Another example — I ask Cursor to improve the system prompt for user profile generation. Within minutes of accepting its suggestions, the quality of the game’s user profiles and dialogue noticeably improved!

I’m convinced AI-driven coding is wildly powerful for:

  • rapid prototyping

  • building minimum viable products (MVPs)

  • iterating fast to achieve SaaS product/market fit

My Favorite Part

Here’s my favorite part of AI-driven coding:

AI handles the repetitive, low-level, plumbing/scaffolding technical tasks, freeing me up to focus on the creative and fun parts.

Instead of troubleshooting bugs (aka reading Stackoverflow all day)…

Now, I can focus on core gameplay, such as generating unique character profiles and ensuring entertaining vampiric dialogue.

Deployment

As the last step I briefly show in my Youtube video, I deploy Vampire Security Checkpoint to Vercel.

I ask Cursor AI to walk me through it step-by-step:

Everything works smoothly.

In minutes, my game is deployed and live at:

Concluding Thoughts

In conclusion, this experiment truly surprised me.

Leveraging AI tools, Claude and Cursor AI, I successfully built a fully functional text-based simulation game in 1 hour.

Without touching a line of code.

Just by talking.

This would’ve been UNBELIEVABLE 10 years ago!

Let that sink in…

Did I miss anything?

Have ideas or suggestions?

Message me on LinkedIn👋

Sabrina Ramonov

P.S. If you’re enjoying my free newsletter, it’d mean the world to me if you share it with others. My newsletter just launched, every single referral helps. Thank you!