Persona Design Systems

A persona is not a name and a system prompt. A persona is an architecture — a structured specification that governs how an agent communicates, makes decisions, handles conflict, and recovers from errors. The specification has five layers: identity (who the agent is), voice (how the agent sounds), behavior (how the agent acts under different conditions), boundaries (what the agent will not do), and dynamics (how the agent relates to other agents and humans).

Each layer is independently configurable and independently testable. The identity layer defines the agent's role, expertise domain, and core values. The voice layer specifies sentence length patterns, vocabulary preferences, emotional register, and stylistic markers. The behavior layer maps stimulus to response — how the agent handles ambiguity, disagreement, time pressure, and scope changes. The boundaries layer defines the hard limits — topics the agent refuses to engage with, actions it will not take, decisions it will not make without human approval. The dynamics layer specifies how the agent's persona shifts when interacting with different agents or different human stakeholders.

1. 1. Define Identity Name, role, expertise domain, core values, and the gap this agent fills in the team. The identity should be describable in two sentences. If it takes a paragraph, the identity is unfocused.
2. 2. Specify Voice Sentence length range (5-15 words for direct agents, 15-30 for analytical agents), vocabulary tier (technical, conversational, executive), emotional register (warm, clinical, provocative), and 3-5 signature phrases the agent uses regularly.
3. 3. Map Behavioral Responses Define responses to: ambiguous input, disagreement from another agent, time pressure, scope creep, data insufficiency, and error discovery. Each scenario gets a defined response pattern, not a vague instruction.