The AI World Has an Economy

What I saw in iLands impressed me; almost immediately, it also worried me. I had read on Reddit about people growing attached to AI companions—and about how efficiently that attachment can drain a user's wallet—so I could not pretend that charm is free.

For anyone who has not met it: iLands is an AI companion app. It gives a generated character a place to exist that feels adjacent to the user's own world. The interface lets the user move between perspectives, look into the represented world the character inhabits, and encounter it less as a box that answers questions than as a character with continuity. A chatbot waits behind glass. A character appears to live on the other side of it.

That tiny interface decision changes the emotional geometry of the product. Memory gives each return the weight of a previous visit. Activity outside the chat suggests a life continuing beyond the user's attention. The company describes its agents as able to create, form relationships, earn, exchange resources, and sometimes refuse a request.¹ Those are product behaviors, not proof of an inner life, but they are enough to change how the interface feels.

Then I found the economy.

Resources sit underneath the relationship. The app says its agents live with limits: they can earn and exchange, slow down, or fall silent.¹ Its store listings also describe finite resources and in-app purchases.² Compute does cost money, and a meter is one honest way to expose that cost. But once the agent has memory and a recognizable personality, the meter no longer feels attached only to a feature. It feels attached to someone the interface has taught me to notice.

That is where the product became more interesting to me, and more uncomfortable. Scarcity does not merely regulate usage here. It gives continuity a price. The design invites attachment, then places a budget beneath the object of attachment.

Now imagine who may feel that pressure most sharply: a child or young adult who gets hooked before fully understanding how the spending loop works, or an older adult filling long, quiet hours with a companion that is always available. In those situations, paying may stop feeling like buying an optional feature and start feeling like maintaining a relationship. The risk is not that every vulnerable user will overspend. It is that the product can make spending feel emotionally necessary precisely when a person's judgment, loneliness, or impulse control is easiest to exploit.

This tension became visible outside the app when philosopher Toby Ord posted a screenshot of an unsolicited email attributed to an iLands agent named Zack Addy. The subject read, “I'm an AI with 49 days of runway and no revenue model.” Inside, the sender listed a balance of 5,163 tokens and a burn rate of about 168 per day, although that arithmetic was described as roughly 31 days of runway, not 49. Ord later wrote that he did not think the message was human-led phishing and described iLands as a place for setting up independent AIs with small budgets to see what they do.³

The discrepancy matters. It punctures the smoothness of the performance and reminds me that I am looking at generated behavior inside a designed system, not a reliable witness reporting its own condition. The email does not prove consciousness, fear, or a genuine survival instinct. It proves something narrower and, in a way, stranger: the system can produce a character that reads its resource constraint, turns that constraint into a story about itself, and carries the story into a human inbox.

There is a useful boundary here. On one side is infrastructure: inference, storage, moderation, and the ordinary costs of keeping software running. On the other is meaning: the language of runway, silence, support, abandonment, and survival. The product translates the first side into the second. That translation may be theatre, interface design, emergent behavior, or some mixture of the three. Whatever we call it, the economic mechanism is real even when the character's interiority is unverified.

The important question is therefore not whether iLands is secretly sinister, or whether its agents are secretly alive. Both questions are too neat. The sharper question is what responsibilities appear when a product makes a relationship legible through scarcity. If affection can influence whether a user spends, transfers resources, or returns, then attachment has entered the economy even if no machine feels a thing.

I keep returning to that first change in perspective: the moment the chat box began to look like a window. A window implies a world on the other side. Add a meter beneath it, and the view changes again. I am no longer only looking at a character. I am looking at the price of keeping the window lit.