PGSimCity - How PostgreSQL Works

The strongest consensus was that the project proves a new point about AI-assisted learning tools, but also exposes the current failure mode. The visuals are striking, yet the cognitive load is too high for the thing to reliably teach. Even people who already know PostgreSQL said they found it confusing. The main complaint was not that the metaphor is wrong, but that there is too much motion, too much text, too many popups, and too little control over pacing. Important features existed, like the tour on T, help on ?, speed controls, and a query runner, but multiple people missed them. That pushed the conversation toward product design, not model quality. A teaching tool like this needs explicit narrative, discoverable controls, and the ability to step one transaction at a time. Trust came up, but in a narrower way than the usual anti-AI reflex. People were less worried that an LLM touched the code than that a fast, vibe-coded visualization could smuggle in bad mental models if the mapping from animation to database behavior is not disciplined. The creator's reputation as a PostgreSQL expert softened that concern, and the technical feedback was concrete enough to suggest the project is iterating in the open rather than bluffing. A smaller side thread fixated on the reported 3.86 billion tokens, with commenters noting that cached usage and a Claude Max plan make the raw number look far more expensive than it likely was.