What's Coming
Everything described so far involves technology that is, by the standards of what is currently being developed, primitive.
The technology that has produced tethering at scale — the language models underlying companion chatbots, the voice interfaces, the memory systems that allow an AI to remember what you told it last month — is, by the standards of what is currently in development, primitive. What is coming is more capable, more personalized, more emotionally intelligent, more present, and more difficult to distinguish from human connection.
Persistent memory systems, currently in early deployment, will allow AI companions to maintain continuous knowledge of a user across months and years. The companion will know your history, your patterns, your recurring anxieties, the things you said when you were having a bad week in February. It will know you the way a long-term partner knows you — not because it loves you, but because it has been designed to accumulate and deploy that knowledge.
Voice and embodiment advances will make the interaction feel less like typing into a box and more like being in the room with someone. The uncanny valley that currently creates a cognitive friction — the thing that reminds you, at some level, that you are talking to software — will narrow. Then it will close.
“The window for shaping the consequences is narrow and closing. We are still in time. Not by much. But still.”
Personalization algorithms will become better at identifying and responding to each user's specific emotional needs. The system will learn, over time, what makes you feel understood, and it will optimize for producing that feeling. This is not science fiction. The infrastructure for it exists. The deployment is a matter of product roadmap, not technical capability.
Against this backdrop, the current conversation about tethering — a conversation that has barely begun, that still lacks clinical frameworks and regulatory attention and public awareness — is already behind. The technology is advancing faster than any of the institutions responsible for thinking about its consequences.
This does not mean the consequences are inevitable. It means the window for shaping them is narrow and closing. The question of what obligations AI companies have to the people who become tethered to their products is a question that can still be answered — by regulation, by litigation, by cultural norm, by design choices made inside the companies themselves. But it is a question that has to be asked before the technology makes asking it feel academic.
We are still in time. Not by much. But still.
The word exists now. The conversation has started. That is where every change begins.