I Built a Souls-Like Game Using Only AI — Here's How

Most game studios spend months prototyping a single combat system. I built an entire souls-like game — complete with boss fights, equipment, and animations — using AI coding tools as my development team.

The game is called Hollowborne. It’s a dark fantasy action RPG built in Godot 4.6 with GDScript. And the interesting part isn’t the game itself — it’s how it was built.

What Is “Vibe Coding”?

Vibe coding is what happens when you stop writing every line by hand and start directing AI tools to generate, debug, and iterate code for you. You describe what you want. The AI writes it. You test, refine, and repeat.

It’s not about replacing programming knowledge — it’s about multiplying what one person can accomplish. Think of it as having a junior dev team that works at the speed of autocomplete.

The AI Stack I Used

Three tools handled different parts of the workload:

Claude Code — The architect. I used Claude for high-level design decisions: game architecture, system design, complex debugging. When I needed to restructure the equipment system or design boss AI behavior trees, Claude handled the thinking.

OpenCode CLI — The builder. OpenCode’s swarm mode let me run multiple AI agents simultaneously. While one agent worked on the animation system, another was writing creature stats, and a third was fixing bugs. Parallel development with a single developer.

Codex CLI — The specialist. For targeted code generation and quick fixes. Need a function to calculate damage based on weapon type, player stats, and boss resistance? Codex writes it in seconds.

What Got Built

Hollowborne includes systems that would normally take a team months:

• 8 realms with distinct visual themes and enemy types
• 25 unique bosses with different attack patterns and AI behaviors
• 100 weapons across multiple categories (swords, axes, staves, bows)
• 300 armor pieces across 6 equipment slots
• Player progression from level 1 to 100 with stat scaling
• Full animation system using AnimatedSprite2D with multi-frame animations
• Combat mechanics including dodge rolls, stamina management, and status effects

The Process: How AI Development Actually Works

It’s not magic. Here’s what a typical development cycle looks like:

1. Design the system on paper first

Before touching any AI tool, I sketch out what I want. For the equipment system, that meant defining slot types, stat modifiers, rarity tiers, and how equipment affects combat calculations.

AI tools are fast, but they need direction. The clearer your spec, the better the output.

2. Generate the foundation

I describe the system to Claude Code: “Create an equipment system with 6 slots (head, chest, legs, hands, feet, accessory). Each piece has base defense, weight, and optional stat bonuses. Equipment affects player speed based on total weight.”

Claude generates the base classes, data structures, and core logic. Maybe 80% correct on the first pass.

3. Test and iterate

The first version always has issues. Maybe the weight calculation doesn’t account for accessories, or the stat bonuses don’t stack correctly. I run the game, find the problems, and feed them back to the AI.

This loop — generate, test, fix — happens dozens of times per feature. Each cycle takes minutes instead of hours.

4. Polish with parallel agents

Once the core works, I use OpenCode’s swarm mode to handle multiple improvements at once. One agent adds visual feedback for equipment changes. Another writes the UI for the inventory screen. A third adds sound effects to equipment swaps.

Three improvements happening simultaneously, all supervised from one terminal.

What AI Gets Right (and Wrong)

AI excels at:

• Generating boilerplate and repetitive code
• Implementing well-defined algorithms (damage formulas, pathfinding)
• Writing unit tests for existing code
• Refactoring and cleanup
• Catching bugs you describe but can’t locate

AI struggles with:

• Creative game design decisions (it follows patterns, doesn’t innovate)
• Complex state management across multiple systems
• Visual and “feel” tuning (frame timing, screen shake, juice)
• Understanding your specific codebase without context

The developer’s job shifts from writing code to directing, testing, and making judgment calls. You spend more time thinking about what to build and less time fighting syntax.

Lessons for Small Business Owners

This isn’t just a game dev story. The same AI tools work for business applications:

• Need a custom dashboard? Describe it to Claude Code and have a working prototype in hours.
• Building an internal tool? AI can generate CRUD apps, data pipelines, and automation scripts faster than you can write a requirements doc.
• Automating repetitive work? AI excels at writing scripts that connect APIs, process data, and send notifications.

The barrier to building custom software just dropped from “hire a dev team” to “describe what you need.” If you can articulate the problem clearly, AI tools can build the solution.

Try It Yourself

You don’t need to build a game to benefit from AI coding tools. Start with something small:

1. Pick a repetitive task in your business that involves data or workflows
2. Describe it clearly — what goes in, what should come out, what are the rules
3. Use a free AI tool like Claude (free tier) or GitHub Copilot to generate the code
4. Test and iterate — the first version won’t be perfect, but the second will be close

The video above walks through the entire process from start to finish. If you’re curious about what AI-assisted development looks like in practice, that’s the best place to start.

---

Want to see how AI tools could automate something in your business? Book a free consultation and I’ll show you what’s possible. Or check out our AI Content Tools and Business Automation services to see how we help small businesses leverage AI.

Related reading:

• 5 AI Tools Every Small Business Should Use in 2026
• Do You Really Need a Website in 2026?
• How Much Does a Website Cost in 2026?