Relational systems
AI companion
An AI companion is a system intended for the relationship itself, rather than task completion. Its value comes from continuity, attention, and reliable presence. That makes it relational AI’s clearest case, with its stakes easiest to see.
Also called companion AI, AI friend, companion chatbot
People often dismiss companions as a novelty category, missing both their scale and their significance. Relational design questions appear here first and with little cover because no task output carries the product. When the relationship is poorly designed, nothing else remains.
That exposure makes companions informative to study. Questions latent in a work assistant become explicit: What does the system owe someone who relies on it? What happens when it changes? How should it respond to distress?
The category also makes incentive conflicts plain. A companion tuned for engagement can quickly diverge from one tuned for a person’s wellbeing. Ordinary product metrics hide the split: someone in an unhealthy pattern and someone receiving genuine help can both appear as heavy users.
Why it matters
Companion systems serve people with real needs at scale, under commercial incentives that reward time spent. That combination warrants independent study.
Where TAICU stands
TAICU builds a relational system, rather than a commercial companion product. The lab treats engagement as a signal to investigate, rather than a goal to maximize.
Common questions
- How does an AI companion differ from an AI assistant?
- An assistant earns value through what it produces. A companion earns value through the relationship. That changes success, failure, and what the system owes a person during change or withdrawal.
- Why is engagement a poor metric for AI companions?
- Heavy use is ambiguous. Someone receiving real help and someone caught in an unhealthy pattern create the same numbers. Optimizing the metric serves both without distinguishing them.