AI Constellation Engineering

Module 2: Agent Identity Design

Learning Objective

Design an agent identity that shapes how the agent thinks, not just what it does. Understand the six components of identity as cognitive architecture and why each one matters.

Module 2 of 12 2 hours Prerequisites: Module 1 (The Architecture Mindset) 45 min lesson + 75 min exercise

The Six Components of Agent Identity

Agent identity is not a paragraph of text. It's a structure with six distinct components, each one serving a specific cognitive function. Skip any one of them and the identity has a gap that will show up in production.

Let's walk through all six. For each one, I'll explain what it does, why it matters, and what happens when it's missing.

Component 1: The Name

An agent needs a name. Not a label like "Agent 1" or a description like "Research Assistant." A proper name with weight behind it.

This seems cosmetic. It isn't. Naming does something specific to cognition, both the agent's and yours. A named agent maintains character consistency across long sessions and complex tasks. An unnamed agent drifts. You've seen this if you've worked with AI for any length of time. The longer the conversation, the more a generic agent's behavior wanders from its original instructions. A named agent has an identity anchor to return to. "I am Sentinel. What would Sentinel do here?" is a stronger cognitive hook than "I am the risk assessment agent."

Naming also changes how you, the architect, think about the agent. When your agent has a name, you start thinking of it as a specific entity with specific characteristics. You notice when it behaves out of character. You refine its identity with more precision. "The research agent should be more thorough" is vague. "Cartwright is hedging too much on low-confidence sources when the brief asks for comprehensive coverage" is specific, actionable, and reveals a design insight.

How to choose a name. Pick a name that carries cognitive associations relevant to the agent's function. Historical figures, literary characters, cultural archetypes. The name should evoke a way of thinking, not just a domain. If you're building a meticulous quality reviewer, naming it after a famously demanding editor or critic creates an implicit cognitive standard that persists through every interaction.

What happens without it. The agent behaves generically. Across long sessions, its cognitive posture shifts based on the most recent input rather than its designed identity. You lose character consistency, and with it, you lose the predictable behavior that makes multi-agent systems work.

The Metaphor