The "Vibe Coding" Revolution: How AI Created Hexaera and Is Transforming Game Development

Digital game development is undergoing an accelerated metamorphosis thanks to advancements in generative artificial intelligence. What once required entire teams of programmers, designers, and testers is now beginning to be prototyped—and even finalized—by individuals armed only with ideas and language models. A new trend dubbed "vibe coding" proposes that the developer act more like a creative director, guiding the AI through natural language, while the machine handles the syntax and code architecture. The recent case of the game Hexaera, created by the Brazilian developer Jhonny and documented in October 2026, serves as a practical milestone for this new technological era.

The Emergence of Hexaera: From an Experiment to a Full 3D Game

The project began with a simple question: how far is it possible to develop a 3D game using exclusively vibe coding with the Claude Code tool? The result exceeded expectations and culminated in Hexaera, a turn-based hexagonal board city-building game.

The title, which can be played directly in the browser (on both desktops and mobile devices) without the need for installation, puts the player in control of a nascent civilization. Starting only with a campfire and seven unexplored hexagons, the user must evolve their society through 7 historical eras:

• Primitive
• Bronze
• Iron
• Medieval
• Industrial
• Modern
• Contemporary

The ultimate goal is to reach the contemporary era and build the "Smart City" monument. To achieve this, the AI did not limit itself to writing basic functions. Claude Code was responsible for creating and maintaining a Game Design Document (GDD) within an Obsidian vault. It also implemented rules, interfaces, deployment infrastructure, and the 3D board itself.

The visual engine was built with the React Three Fiber library, adopting a voxel diorama aesthetic. Visual elements such as crops swaying in the wind, animals walking across the map, and active chimneys were all generated by the AI. Complex models (such as nuclear power plants and railroads) were modeled programmatically, and procedural chiptune-style audio was implemented via code, eliminating the need for traditional external sound files.

Automated Playtests: AI Testing AI

One of the great differentiators in the development of Hexaera was the approach to solving balancing issues. Strategy games require endless hours of testing to avoid broken mechanics. Instead of playing manually hundreds of times, Jhonny asked Claude Code to create a headless bot (without a graphical interface) to run matches in the background.

The result was revealing. The bot identified critical flaws in the game design, such as:

• Deadlock between steel and energy: Production chains that would lock each other out, making player progress impossible.
• Hunger spiral: An unrecoverable domino effect generated by drops in food production.
• Pacing issues: Some eras lasted much longer than others, breaking the fluidity of the campaign.

After 8 rounds of adjustments guided by this data, the automated bot's win rate jumped from 0% to approximately 67%. Today, a full campaign in Hexaera takes about 290 turns, proving that AI can not only write code but also act as an efficient Quality Assurance (QA) team.

Context and Background: A Global Creation Movement

Hexaera is not an isolated case. The year 2026 was marked by a flood of successful experiments using Claude Code and similar platforms for game development.

In March 2026, Traffic Architect emerged, a 3D browser-based traffic management simulator built with Claude Code and Three.js. The AI generated everything from the camera system to pathfinding (A*) algorithms for vehicle routing, leaving only the pacing and difficulty decisions to the human.

Shortly after, in June, developer Lucas Kazansky presented a prototype for an AR City Builder (an Augmented Reality city builder) that did not require a single line of manual code. Using Claude Code, Coplay MCP, and Unity, the project demonstrated how the transition from ideation to functional prototype became almost instantaneous.

In September, the community witnessed an even more extreme milestone: using the newly released Claude Fable 5.1 model, a Reddit user generated a prototype with the complexity of Cities: Skylines II from a single initial prompt and an empty folder. The operation took an hour and a half and consumed US$ 203 in API costs. The secret to this experiment was not asking for the game directly, but rather instructing the AI to first create an architecture (in an ARCHITECTURE.md file) and establish its own validation methods.

However, vibe coding is not without its pitfalls. Developer Cristian Trucco, who in August 2026 published the game NOVA WING — Star Assault (a 3D rail shooter inspired by Starfox with over 4,600 lines of code and 16 3D models), reported that the challenge is not generating code, but managing complexity. In a single session, he lost eight validated fixes because the source code and the published file lost synchronization. Trucco’s account highlights that, no matter how advanced the AI is, classical software engineering disciplines—version control, reliable testing, and bug investigation—remain fundamental.

Impact on the Sector, Users, and the Latin American Market

The adoption of these tools represents a colossal impact for the technology and entertainment industry, especially for independent developers and small studios.

For the sector, it means a drastic reduction in the barrier to entry. In the past, creating a 3D game with physics mechanics, dynamic economy, and procedural audio required formidable capital and years of studying programming languages. Today, the role of the developer is elevated to that of a systems architect and experience designer.

For users, this democratization promises an explosion in the variety of niche games. When production costs fall sharply, it becomes viable to create hyper-specific games that wouldn't need to sell millions of copies to be financially justified.

In the Latin American market, the impact is particularly transformative. Brazil, which already has a massive gaming consumer market, often struggles with a lack of funding for local studios. Projects like those by Brazilians Jhonny (with Hexaera shared via TabNews) and Cristian Trucco show that creators in Latin America are at the forefront of this technological adoption. With an AI subscription or a few dollars in API consumption, creative minds from the Global South can now compete in technical complexity with larger teams from traditional markets.

What to Expect Next

The current scenario suggests clear trends for the coming years in software and game development:

• Autonomous QA Agents: The use of bots integrated into code to test thousands of matches simultaneously will become an industry standard, solving complex mathematical balancing problems.
• Copiar
  Sync and Versioning Tools for AI: As warned by cases where code was lost, we will see the emergence of specialized platforms dedicated to maintaining version control strictly focused on prompts and "vibe coding," preventing AI from overwriting validated solutions.
• Real-time Asset Generation: In addition to generating programmatic logic, upcoming models should be capable of generating textures, orchestrated soundtracks, and high-fidelity 3D models on the fly, replacing the voxel visual style (used in Hexaera) with photorealistic or highly stylized graphics on demand.

Frequently Asked Questions

What is "vibe coding"? It is a software development approach where the creator focuses on describing the architecture, business rules, and desired behavior using natural language, while Artificial Intelligence (such as Claude Code) is responsible for writing, structuring, and deploying the code itself.

Can anyone play Hexaera? Is there a cost? Yes, Hexaera is completely free and does not require installation or account creation. It runs directly in the web browser and is compatible with both computers and mobile devices.

How does the playtest bot created for Hexaera work? It is a robot that plays the headless version (without a graphical interface or 3D graphics) of the game at high speed. It executes hundreds of matches simulating player decisions to discover mathematical flaws, excessive difficulty, and unbalanced resources, generating improvement reports for the developer.