Context is identity

Every AI conversation starts from zero. Every human conversation doesn't.

• My daughters know how I laugh.
• My friends know what frustrates me.
• My wife knows when I'm quiet for too long.

Every meaningful human relationship is built on accumulated context, and AI starts every one as if we've never met.

Yet the more I worked with AI, the more I found myself asking a different question. Every time I finish a long conversation with AI, I pause before closing the window and ask myself three things:

• what do I own?
• Did I actually think? Or
• did I merely approve what the AI produced?

Those questions led me somewhere I wasn't expecting. I had three opportunities in front of me:

• a high-paying consulting job,
• a chance to teach AI to beginners,
• and a project that aligned with my values but earned little.

So I asked an AI I'd personalized in Notion a specific question:

Based on what you know about me, how would you help me think about this? Answer this but don't give me advice. Ask me only questions until I arrive at my own answer.

It didn't answer. Instead, it asked me twenty-five questions. What surprised me wasn't the quality of the questions. It was that they sounded exactly like questions I would ask.

Intention

I didn't set out to discover this. It happened by accident, and I'm not ashamed to say so. This is me recording the accident for future reference.

Working with AI changes authorship. At first you write and the AI assists. Gradually you become a co-author. Eventually, if you're not careful, you become an editor of thoughts that aren't entirely your own. You stop generating. You start validating.

I'm not a native English speaker, so I work on multiple levels at once: a first pass of language translation, then understanding the response, then reacting to it. If you're a non-native speaker too, you'll know what I mean. AI doesn't simply answer questions. Over time, it begins to influence the way we ask them. That isn't necessarily good or bad, but it deserves attention.

A long AI thread sometimes drifts onto completely new ground. Sometimes that's annoying and cognitively overloading; you have to course-correct the conversation more than once, and it doesn't feel natural. Guardrails tell you where not to go, but they're not directive enough to stop new, unintended ideas from springing up. Sometimes I feel like a passenger in a car driven by AI. That sense of lacking ownership, as a passenger, comes from the fact that the driver doesn't share your identity.

Identity

Originally I thought context was documents. Then context was conversations. Then context was memory. Now I think there's another level:

Context is identity. Identity influences judgment. Judgment influences advice. Therefore identity belongs in the context.

I've run a lot of experiments across different ideas: context orchestration, behavioural contracts, human orchestration, Contour Mind. Most of that earlier work focused on giving AI more context about the task. The shift now is toward giving it more context about me.

If your AI knew what you fear, what you value, how you make decisions, and the principles you refuse to compromise, would it give different advice? Should it?

Same prompt, different human

Suppose two people ask the same model the same question: should I quit my job? Same prompt, same model, completely different identity. The answers should differ, not because of demographics, but because of values and identity. Neither answer is universally correct. They should be aligned with the individual's own identity.

If identity changes the conversation, then maybe identity deserves to become part of the context we give AI. That led me to a simple idea: what if every person had a Personal Constitution?

Personal Constitution

A Personal Constitution is a portable document that lets an AI know your values, your principles, your decision framework, and your preferred way of reasoning, and helps you remain aligned with them while still being challenged when appropriate. It tells the AI not who I am, but how I choose to live.

Ironically, the greatest beneficiary of writing mine wasn't the AI. It was me.

A few lines from it, as an example:

• I value contribution over recognition.
• Challenge my assumptions.
• Protect my health.
• Teach me through questions.
• Optimize for alignment.
• Truth before comfort.

None of these are preferences. They're principles.

A minimal constitution

At the bare minimum, a Personal Constitution has three levels:

• My Constitution: who I am.
• My Decision Framework: how I make decisions.
• My Question Framework: how I investigate reality.

What I keep

At the highest level, mine covers who I aspire to become, my principles, my decision framework, what I refuse to compromise, how I learn, how I want AI to challenge me, and what I'm trying to protect.

I intend to revisit it once or twice a year and ask: "Am I still living according to my constitution?" I feel a bit more peace just asking that question.

After an AI conversation now, I want to be able to say I own the thinking, and that I haven't drifted away from it.

AI may eventually know our preferences better than we do. The more important question is whether it understands our principles. Preferences make us convenient. Principles make us who we are.

And perhaps the greatest value of writing a Personal Constitution isn't teaching AI how to help us. It's teaching ourselves who we intend to become.

If an AI remembered only one thing about you forever, what would you want it to be?