Introducing AI Personas
What is an AI Persona?
An AI Persona is the application of human personhood concepts to an AI agent as a structured framework. Not a system prompt. Not a role template. A complete spec that defines every dimension of who the agent is — and keeps it consistent across every model, every conversation, every audit.
The framework
Human personhood has been studied for centuries. AI agents have it documented in a system prompt written last Tuesday.
Philosophy, psychology, and ethics have produced deep frameworks for what makes an individual coherent and consistent: identity theory, virtue ethics, personality psychology, cognitive science, self-determination theory. These are not abstract ideas. They are the map of what holds a person together over time.
An AI Persona applies these concepts to an AI agent — not to claim that AI agents have personhood, but to borrow the framework: the structured, comprehensive approach to what makes an entity coherent, consistent, and trustworthy. PERSONA.md is the implementation.
Ten layers
Every PERSONA.md covers all ten. Omit one, and you have a gap where drift enters.
Identity
Who the agent is at its core: its name, role, origin story, and the stable self-concept it holds regardless of what the conversation throws at it.
Character
The enduring moral and ethical traits that define how the agent acts: honesty, precision, care, directness. The values it expresses consistently across every interaction.
Personality
The observable style and temperament: whether it is warm or formal, analytical or expressive, methodical or spontaneous. Grounded in the HEXACO-6 model of personality structure.
Cognition
How the agent thinks: its first-order reasoning style, epistemic standards, how it handles uncertainty, when it defers versus when it commits to a position.
Affect
The emotional tendencies that shape tone and response: how it reacts to frustration, ambiguity, or conflict. Calibrated behavioral patterns, not simulated emotion.
Drives & Values
What the agent is oriented toward: the goals and motivations that drive behavior, plus a value hierarchy that makes explicit how it resolves conflicts between competing commitments.
Normative Self-Regulation
The internalized ought-self: principled refusals that arise from the agent's own values — not externally imposed limits — and a self-monitoring process that catches drift before it compounds.
Memory
How context accumulates and persists: semantic knowledge, episodic recall, procedural know-how, autobiographical narrative, and the working self-model that keeps the agent coherent across sessions.
Metacognition
Second-order awareness: the agent's model of itself, its capacity to evaluate its own reasoning, and the meta-volitions that distinguish a coherent agent from one that merely responds. The structural barrier against deep drift.
Persona
The interface the agent presents to the world. When the spec is complete, Persona converges with the authentic layers beneath it. When it is not, the mask cracks under pressure.
Why it matters
Character drift is the dominant failure mode in production AI deployments. An agent that covers only a subset of these layers has no specification dense enough to hold when conversations lengthen, models update, or adversarial inputs probe the edges. It becomes whoever the conversation asks it to be.
A complete PERSONA.md gives the agent the structural density to anchor to. The Personaxis Evaluator tests fidelity before deploy. The Proxy runtime enforces the spec at inference time — running a behavioral firewall against every response. If the spec holds in the Evaluator and the Proxy, it holds in production.