Detailed Analysis
A Reddit user has successfully developed a fully playable turn-based RPG titled *Teotlan: Land of Gods* using Claude as a collaborative coding partner, publishing the completed game directly through Claude's public artifacts feature. The game draws on Mesoamerican and Aztec mythology, casting players as deities navigating nine layers of Mictlan, the Aztec underworld, with roguelite progression mechanics, 12 playable gods, and systems for capturing, sacrificing, and summoning units. The developer reached the finished product through more than 170 iterative versions, a figure that underscores both the complexity of the undertaking and the sustained back-and-forth dialogue required to produce a coherent, bug-free experience through AI-assisted development.
The developer's methodology reveals a deliberate, structured approach to AI-assisted coding that stands in contrast to more spontaneous "vibe-coding" workflows. By front-loading a formal design document and locking down game logic before any code was written, the developer gave Claude a consistent reference framework, which served the dual purpose of grounding Claude's outputs and providing a benchmark against which hallucinations or logical inconsistencies could be identified and corrected. This disciplined process — write spec, generate build, playtest exhaustively, document findings, iterate — effectively treats Claude less as an autonomous developer and more as a highly capable executor operating within human-defined constraints. The developer's explicit note that a prior attempt on ChatGPT failed suggests that the particular qualities of Claude's code generation or instruction-following made a meaningful practical difference in this use case.
The publication of the game via Claude's public artifacts system is itself a notable detail. Claude's artifacts feature allows users to share interactive, self-contained outputs — including functional web-based applications and games — directly through a Claude-hosted URL without requiring external hosting infrastructure. For individual developers without deep deployment expertise, this significantly lowers the barrier between creation and distribution, collapsing what would traditionally be a multi-stage release process into a single shareable link. The fact that a feature-complete RPG with turn-based combat, multiple character classes, roguelite progression, and unlockable content can be hosted and played through this mechanism illustrates how far the capabilities of in-context AI tools have extended into domains previously requiring specialized teams.
This project sits within a broader and rapidly accelerating trend of non-professional developers using large language models to build software that would historically have demanded years of programming expertise. The roguelite and RPG genres, while beloved, are mechanically demanding — requiring state management, combat resolution logic, branching progression systems, and UI coherence — making this a more technically ambitious example of AI-assisted development than typical demonstrations. The 170-version iteration count also speaks to a broader truth about this development paradigm: the process remains highly iterative and human-supervised, with AI accelerating execution rather than replacing the design intelligence, creative vision, or quality assurance judgment of the human collaborator. As tools like Claude continue to improve and artifacts-style deployment becomes more sophisticated, this kind of solo-developed, mythology-rich interactive experience may become increasingly representative of what individual creators can produce outside traditional studio structures.
Read original article →