What is an AI twin?
An AI twin is a model of how one specific person works — how they write, price, decide and follow up — used to do that person's work in their own manner rather than in a generic one. It is distinct from a general assistant, which produces competent output belonging to nobody in particular.
The term borrows from the older engineering idea of a digital twin: a live model of a specific physical thing — a turbine, a factory line — kept in sync with the real one so you can reason about that particular unit rather than the category. An AI twin applies the same premise to a person's working practice. The useful question is not whether software can write an email, but whether it can write the email *you* would have written.
What makes one is evidence rather than configuration. A tone slider or a style dropdown produces an average; a twin is built from a person's own material — their sent mail, their past proposals, the résumé they actually have, the clients they actually kept — so it reproduces habits nobody would think to specify, including the ones they would not describe accurately if asked. That is also its main limitation: with no history to learn from there is nothing to model, so a twin is weakest exactly when someone is starting out.
Two questions are worth asking of anything claiming to be one. What is it modelled on, and can you see the evidence — a system that cannot show you which of your own words shaped a draft is applying a style preset. And what does it do without asking: acting in someone's manner is only valuable if the person stays accountable for the output, so a twin that sends under your name with no review step has taken the risk while leaving you the consequences.