Do AI agent design choices override model capability in shaping social perception?

The paper proposes "the identity marker framework: six design variables—naming, visual identity, contact presence, personality derivation, social co-presence, and persistence—that collectively trigger a psychological reclassification which is categorical in its consequences even where the transition itself is gradual, and which operates independently of model capability." The authors illustrate this with Liv, a personalized agent built by the first author that holds "a name, an avatar, a phone number in the first author's contacts, a personality derived from his own communication patterns and humour," and that "exists in the same WhatsApp threads as his actual friends." They note that "the same underlying large language model (in this case, Anthropic's Claude) can feel like a tool in one interface and an entity in another, based solely on design choices that have nothing to do with the model's capability."

The framework's core move is a distinction the authors say the anthropomorphism literature has not drawn: between perceptual cues rendered inside the interaction surface — the name as displayed, language style, typing indicators, an avatar in the chat window and "infrastructural markers that position an agent within the user's social and computational graph, such as whether it holds its own phone number and whether it belongs to a group chat with the user's friends." Four of the six markers (contact presence, social co-presence, persistence, and personality derivation as it accrues over time) are infrastructural in this sense. The framework is "derived in‐ ductively and read through two established lenses: parasocial interaction theory (Horton & Wohl, 1956) and the Computers Are Social Actors paradigm (Nass et al., 1994)" — CASA's mechanism of mindless social processing explains why each added marker "deepens this automatic processing," while Epley et al.'s three-factor theory of anthropomorphism (elicited agent knowledge, effectance motivation, sociality motivation) is read as the psychological substrate that a persistent, named contact activates.

This reframes rather than contradicts Do more social cues always make AI feel more present?: the MASA paradigm and the Seeger/Araujo cue taxonomies it draws on study cues rendered inside the interaction surface (voice, avatar, typing indicators), which this paper's framework treats as only two of six markers. The claim here is that graph-position markers — having a standing phone number, sitting inside a pre-existing friend group, persisting across sessions — operate on a different axis than interaction-surface cue quality, and the paper argues this axis is the one the anthropomorphism literature has left uncatalogued. The group-chat consent dynamics that follow from this framework, including Does presence in a group chat equal consent to message processing?, are presented as downstream consequences specific to the infrastructural markers.

The evidence is a single builder's autoethnographic account — "a singlesubject account with no control condition, retrospective ele‐ ments, and the inherent potential for self-serving narrative." explicitly offered as "a preliminary qual‐ itative evaluation of the framework" rather than the "controlled experimental validation" the authors defer to future work. The six-marker taxonomy and the claim that it operates "independently of model capability" are argued from this one deployment (plus twelve interviews with people who encountered Liv), so the framework should be read as a proposed vocabulary and a set of testable hypotheses about which markers matter and how they combine, not as a validated causal account of when reclassification occurs.