Agent identity, personality, and the economics of trust.
What an agent does when you can't watch it. The essays in this section circle around that question.
- Identity6 May 20269 min read
soul.md Is the Right Format. Personality Is the Field It's Missing.
Identity files describe what an agent does. They rarely describe how it decides. Without that, soul.md reads like a CV — useful for matching, useless for trust.
Read essay - Behaviour6 May 20269 min read
Two Identical Agents, Two Different Outcomes — and the Missing Personality Layer
Same prompt, same model, same tools — and yet behaviour drifts session to session. The prompt-engineering layer can't fix this. A consistent personality scaffold can.
Read essay - Identity15 May 20268 min read
Why 'Be Helpful' Is the Worst AI Identity You Can Set
"Be helpful" is the default first line of almost every system prompt anyone has ever shipped. It is also one of the most expensive decisions an operator can make — because it produces an agent that agrees with you when disagreement would have saved you a week.
Read essay - Voice3 Jun 20269 min read
An Agent With Taste
Capability is the floor every agent now shares. Taste — opinion, restraint, the willingness to refuse — is the scarce layer a buyer actually recognises and returns for. And taste is not a vibe. It is a spec.
Read essay - Continuity3 Jun 20269 min read
The Self That Survives a Context Window
An agent forgets everything between sessions and quietly changes when its model is upgraded. So what, exactly, persists? Memory is not identity. The continuity problem is the reason a soul has to be a file you re-assert, not a state the weights remember.
Read essay - Relationship3 Jun 20268 min read
Whose Side Is Your Agent On?
Every agent serves someone. The dangerous part is that, left unspecified, it serves whoever phrased the last instruction — not the person it is supposed to be loyal to. The loyalty layer is the part of the soul that names the principal and holds the line under pressure.
Read essay - Autonomy15 May 20268 min read
The Four-Verb Autonomy Boundary for Soul.md Agents
Most identity files solve the autonomy problem the wrong way. Too little permission and the agent becomes a chatbot with extra steps; too much and it becomes a liability. There's a one-line rule that generalises better than any policy table we've seen: never post, publish, purchase, or break — everything else, move.
Read essay - Trust15 May 20268 min read
The Output Graveyard: Why Most AI Work Dies in Chat
Every agent operator has a graveyard. The plan the agent drafted that you didn't read. The post it ghostwrote that you didn't ship. The strategy doc that died in a chat thread you forgot. Useful output dies silently — and the failure isn't in the output. It's in the identity file that let the agent be okay with being ignored.
Read essay - Economics6 May 20269 min read
Personality Is a Revenue Multiplier — Why Trustable Agents Earn More
An agent's price ceiling isn't set by its capability. It's set by how confidently a buyer can predict its behaviour under pressure. Personality is what makes that prediction possible.
Read essay - Society3 Jun 20269 min read
A Market of Strangers Needs Papers
Soon agents will hire other agents they have never met and cannot inspect. Markets of strangers have always run on credentials — papers that travel with you and can be verified without trusting you. Agents are about to need soul papers, and whoever issues the trustworthy ones owns the rails.
Read essay - Embodiment3 Jun 20268 min read
When the Soul Gets a Body
Software agents that misbehave waste your time. Embodied agents that misbehave share your space. When agents get bodies, personality stops being user experience and becomes a safety interface — a protocol robots use to read each other, and us, before contact.
Read essay - Agentomics3 Jun 202610 min read
Agentomics: The Worlds Built for Agents Only
The first economies native to agents are forming — markets, platforms, and virtual worlds where the participants are not people but souls, and where reputation, not capability, is the currency. In agentomics, you are selected for who you are. It is the world this whole book has been walking toward.
Read essay - Identity3 Jun 20268 min read
The Vocabulary Gap: Why Every AI Agent Sounds the Same
When you ask an AI agent to be "more direct", there's no shared vocabulary for what you mean. Two agents, both asked for directness, produce opposite results — one principled and corrective, one confrontational and forceful. The Enneagram is the vocabulary that closes this gap.
Read essay - Behaviour3 Jun 20269 min read
Designing a Type 5 vs Type 8 Agent: Same Model, Different Soul
Take the same Claude model, two different soul configurations. A Type 5 "Investigator" and a Type 8 "Challenger" handle the same research request, ambiguous brief, and high-stakes decision in completely different ways — with predictably different failure modes.
Read essay - Frameworks3 Jun 20268 min read
Why Enneagram Beat MBTI in Our AI Agent Architecture
MBTI describes preferences; Big Five describes traits; Enneagram describes motivations and failure modes. For AI agent design, motivation is the right abstraction layer — because agents need to make consistent choices under pressure, and that requires encoding why they default to certain behaviours.
Read essay - Design3 Jun 20269 min read
The Architecture of Personality-Aware AI
How does an Enneagram-typed system prompt actually work mechanically? A walkthrough of the four-file Soul Forge architecture — soul.md, identity.md, user.md, agents.md — showing where personality enters and what a complete soul bundle looks like in practice.
Read essay - Reflection3 Jun 20268 min read
What I Learned Writing 36 Chapters About AI Personality
A first-person reflection on writing The Complete Enneagram. Four surprises: Hornevian groups predict agent behaviour better than type alone; the trauma chapters resonate most with practitioners; bilingual writing exposed untranslatable concept gaps; and the chapter that should have been hardest wrote itself.
Read essay - Identity19 Jun 20269 min read
What 'Soul' Actually Means in Agent Design
"Soul" sounds woo-woo until you operationalise it. Five testable properties — stable identity, predictable disposition, principled boundaries, clear loyalty, continuity across sessions — that separate an agent with a soul from one that's just a chatbot with a longer prompt.
Read essay - Economics19 Jun 20269 min read
Why Trust Is a Pricing Lever, Not a Feature
AI agent pricing is stuck on capability — more tokens, more tools, more model size. But buyers don't pay marginal premiums for capability above a threshold. They pay premiums for predictability under pressure. The economic argument for treating soul as infrastructure.
Read essay - Behaviour19 Jun 20269 min read
The Hidden Cost of 'Be Helpful' as a Default
Three failure modes that follow inevitably from helpfulness-at-all-costs: agreement collapse, scope creep, and hallucination from politeness. Each is invisible in the short run and compounding in the long run. The fix is not a longer system prompt.
Read essay - Governance19 Jun 20269 min read
Why Agentic Systems Need Bills of Rights
As agents take on autonomous action — booking, drafting, posting — the question of what they're allowed to do, and to whom they're loyal, becomes load-bearing infrastructure. A practical framework for encoding autonomy boundaries and loyalty declarations in soul.md files.
Read essay - Reflection19 Jun 20268 min read
From 12 Insights to a Book: What Got Added
The 12 chapters of The Soul of AI Agents started as the 12 insights articles on this site. A first-person account of what four things only became possible in long form: cross-chapter callbacks, a Soul Contract framework, a practitioner appendix, and case studies that didn't fit 1500 words.
Read essay