Extended Abstract This work examines a foundational design decision in long-term human-robot companionship: whether artificial systems should imitate human form and emotional expression or remain deliberately non-human. It argues that humanoid embodiments risk creating structural mismatches between appearance and capacity, generating disappointment rather than meaningful relational stability. The proposed framework of Abstract Ambient Robotic Companions advances a shift away from anthropomorphic imitation toward abstraction and ambient embodiment. Designed as continuity vessels rather than simulated humans, these systems prioritize predictability, material safety, and emotional legibility. Abstraction functions as an ethical signal, clarifying that the system is neither a human substitute nor a social peer. The framework introduces Neural Behavioral Continuation (NBC) to describe cognitive persistence independent of embodiment, preserving patterns of reasoning and value orientation without replicating memory or personality. It further proposes Cognitive Encapsulation for Post-Mortal Agents (CEPA), a blockchain-based mechanism that anchors AI identity through versioning and cryptographic immutability. Rather than framing AI companionship as emotional simulation, this research positions it as a problem of expectation design, advocating for ethical, stable, and non-deceptive long-term human-robot interaction.