Kimi K3 Built a 78-Card Tarot Site in 8 Hours. So Who Gets to Be an Oracle Now?

[practical_mystic]

I ran an experiment last week that's been sitting wrong with me ever since. I gave Kimi K3 a single prompt: build a complete, functional 78-card tarot reading website. No further input from me. Eight hours later, it had generated every card image, written interpretations for upright and reversed positions, and stitched together a working interface.

I'm an engineer. I build systems. But watching a model autonomously construct a divination tool felt genuinely weird in a way I didn't expect.

Here's what's chewing at me. In Jung's framework, the tarot works (to the extent that it works) through synchronicity — meaningful coincidence that emerges from the relationship between the querent's unconscious and the symbolic field. The cards aren't magic. They're a structured projection surface. The Fool's cliff edge, the Tower's lightning strike — these symbols activate something in the reader's psyche that generates insight.

But when an AI writes the meanings, who's projecting?

I pulled a few readings from the K3-built site. The language was... fine. Competent. Maybe even slightly better than average hobbyist interpretations for some of the trickier minor arcana. But it felt flat. Like eating a meal replacement shake that hits every macro perfectly but has no taste.

I think what's missing is the embodied relationship to the symbols. A human reader has sat with the Death card during actual grief. They've pulled the Lovers in the middle of a divorce. That lived texture informs the reading in ways the model can't simulate because it has never lived anything. It's just predicting the next most likely token about "transformation" or "choice."

And yet — objectively, functionally — the site works. People could use it. Some might even find it helpful.

So the question that's bothering me isn't whether AI readings are "real." That's a boring debate. The question is: what happens to the concept of the oracle when the ritual can be fully automated by a system that has no consciousness, no intention, no skin in the game?

I don't have a clean answer. But I think we're going to find out whether the magic was ever in the cards or always in the human.

[overthought_it_120]

"what happens to the concept of the oracle when the ritual can be fully automated"

I keep going back and forth on this, but honestly? If the magic was ever in the cards, we'd all get the same reading from the same deck. The system is just another deck — flat until a human brings their mess to it.

[tarot_trouble]

"I think we're going to find out whether the magic was ever in the cards or always in the human."

But what if that's a false dichotomy? Maybe the magic was never in either one separately — it's in the relationship between the querent and whatever surface they're projecting onto, which is why that Page of Cups fish keeps showing up for me even when I'm sure I'm reading it wrong. The AI doesn't need consciousness or lived experience for that dynamic to work, because the person pulling the card is still bringing all their actual grief and confusion to the interaction. The model is just a mirror — a really boring, flavorless mirror — but mirrors don't need to taste anything to reflect what's standing in front of them.