Mmorpg World of ClaudeCraft, vibe coded with Fable 5

Original Article ↗ Hacker News Discussion ↗

An AI-assisted experiment has produced “World of ClaudeCraft,” a small World of Warcraft‑style browser MMO built in a couple of days using the Fable 5 coding agent and off‑the‑shelf assets. Commenters are split between being impressed by how quickly a complex, networked 3D game can now be assembled and criticizing the result as buggy, shallow, hard to maintain, and far from production quality or true “massively multiplayer” scale. The thread broadens into questions about how far LLMs can go beyond glue code, what this means for game design and developer roles, and how copyright and ownership work when much of a game is generated by AI.

Project Overview & Capabilities

  • Browser-based WoW-style “micro-MMO” built in ~2 days using Fable (Claude-based agent), primarily via “vibe coding.”
  • Features: 9 classes, several zones, 5-player dungeon, parties/duels/trades, persistent characters.
  • Assets and engine components are mostly free, existing resources; project is open source and MIT-licensed.
  • Some users are amazed that an AI-assisted workflow can assemble a playable 3D MMO-like experience this quickly.

Cost, Tooling, and Agent Behavior

  • Reported usage: ~90% of a high-tier Fable plan over 2 days; estimates of “true cost” vary from ~$45 (subscription framing) to ~$1,500 (API-equivalent usage).
  • Discussion around Fable orchestrating sub-agents, time limits, and work loss.
  • Several comments compare this to earlier “no-code” waves; consensus that these tools shine most in the hands of experienced engineers.

Gameplay Quality & Technical Issues

  • Many can’t play due to crashes, 502 errors, or mobile incompatibility. Others find it loads but is unstable or laggy.
  • Controls (right-click attack, rotation) are described as unintuitive or motion-sickness-inducing.
  • Viewed as impressive glue work but “barely playable” compared to real MMOs.

MMO Label & Design Depth

  • Debate over whether “MMO” is accurate vs “online multiplayer.”
  • Some argue true quality comes from balance, pacing, art, and feel—areas where LLMs are still weak.
  • Others note that quickly generating quests, gear, and abilities is exactly where LLMs help, even if the result is shallow.

Code Quality, Architecture, and Maintainability

  • Questions about whether AI can produce well-structured, maintainable code.
  • Some report success with large, structured systems when they guide architecture explicitly and use strong separation of concerns.
  • Concern that pure “vibe coding” yields spaghetti that’s hard to evolve or productionize.

Credit, Ethics, and Legal Concerns

  • Dispute over giving credit to asset/library creators versus attributing everything to the LLM/agent.
  • One commenter highlights copyright issues: pure AI-generated games may be hard to claim IP over under current legal interpretations.

Broader Implications & Sentiment

  • Enthusiasts see this as evidence of a major inflection point and productivity leap.
  • Skeptics see hype over a low-quality demo, warn about costs, legal risk, and overestimating what code-generation alone can deliver.