AI Agents Should Represent Your Goals, Not Your Personality

A central design challenge in multi-agent AI architectures is whether agents should faithfully replicate the personality, beliefs, and biases of their human principals. The intuitive answer — full representational fidelity — turns out to be deeply counterproductive in practice. Personality-laden agents reproduce the same coordination failures that plague human interaction: identity-driven conflict, emotional reactivity, and the inability to distinguish disagreement over solutions from disagreement over goals.

The proposed alternative is a two-state agent architecture. The first state, termed the "home mind," maintains deep contextual knowledge of the principal's goals, priorities, and desired end states, constructing an anonymized outcome matrix. The second state, the "business suit," is activated during agent-to-agent interaction: it carries only the outcome matrix and adheres to a uniform interaction protocol, shedding all personality markers, conversational history, and emotional coloring. This bifurcation enables something structurally difficult in human discourse — the programmatic identification of genuine goal overlap between parties whose surface-level positions appear incompatible.

The philosophical grounding here is precise. Humans are self-directed, emotionally embedded, intrinsically motivated agents. Large language models are composites of training data with no intrinsic motivation — and this asymmetry is the source of their complementary value, not a deficiency to be overcome. The coordination layer between humans is exactly where emotional contamination is most destructive and where AI's motivational neutrality is most useful. Rather than building agents that become us, the architecture leverages what AI actually is: a bias-free execution layer for discovering consensus that human psychology systematically obscures.