Writing — AI Coexistence

Why AI Needs a Constitution, Not Just Guardrails

November 2, 20266 min readBy Milan Cheeks

Guardrails are product patches. A constitution is a shared, lasting set of principles for how humans and AI systems live and work together.

Why AI Needs a Constitution, Not Just Guardrails

Every AI product shipping right now has guardrails. I know, because I build these systems. I set the filters, write the refusal rules, and test the edge cases. Guardrails are necessary. They are also nowhere near enough.

Here is the problem nobody in the product meetings wants to sit with: a guardrail belongs to one product, one company, and one version. When the product changes, the guardrail changes. When the company gets acquired, the guardrail can disappear. When a competitor ships with weaker rules and grows faster, the pressure to loosen your own is enormous. None of that is a foundation you can build a civilization on.

And that is the scale of what we are doing. AI systems are moving out of chat windows and into work, school, money, health, and the operating layer of a person's day. I build toward that world on purpose. SELF OS gives a person an AI layer that helps run their goals and habits. Manifest Anything turns a person's intent into a playable game. These are early versions of a permanent arrangement between humans and intelligent systems.

Permanent arrangements need more than patches. They need principles that outlast any single company. That is what a constitution is.

What guardrails actually are

A guardrail is a rule inside a product. Do not generate this. Refuse that request. Escalate this conversation to a human. These rules are set by the team that ships the product, enforced by that team's code, and changed whenever the roadmap says so.

I am not against them. I use them in my own builds. A system without guardrails is a system nobody should trust, including the person who built it.

But notice what a guardrail cannot do. It cannot tell you what a person is owed when an AI system makes a decision about them. It cannot tell a company what duties it carries when its system becomes part of someone's livelihood. It cannot give two competing platforms a shared floor of behavior that users can count on no matter which logo is on the screen. Guardrails govern products. Nothing about them governs the relationship.

What a constitution does differently

A constitution is not a longer list of rules. It is a different kind of document, and it does three things guardrails cannot.

First, it is durable. It is written to survive products, companies, and news cycles. The principles do not get rewritten every release. They change slowly, deliberately, and in the open, instead of being quietly edited in a settings panel.

Second, it is shared. A constitution is a common reference point that builders, users, and regulators can all point to. When my principles live only inside my company, you have to trust me. When they live in a public document I have committed to, you can hold me to them.

Third, and this is the part most AI safety talk skips, it runs in both directions. It states the duties of AI systems and the companies behind them, the rights of the people living alongside those systems, and the responsibilities people carry in how they direct them. Rules for only one side of a relationship are not a framework. They are a terms of service.

Guardrails tell a product what it cannot do. A constitution tells all of us what we owe each other.

Rights and duties, on both sides

So what belongs in it? I speak on this under the title "AI Coexistence & Constitutional Guardrails," and the structure I keep coming back to is simple: rights and duties for people, and duties for the systems and the companies that field them.

People hold the rights. A person has a right to know when they are dealing with an AI system. A right to understand, in plain language, why a system made a decision that affects them. A right to appeal that decision to a human. A right to their own data, and to leave with it. These are not features. Features get cut. Rights are the floor.

AI systems, and the companies accountable for them, carry the duties. A duty not to deceive. A duty to stay within the authority a person actually granted, not the authority buried in a consent screen. A duty to keep a record of consequential actions so a human can audit them later. A duty to refuse instructions that would harm people who never consented to be part of the interaction.

And people carry duties too. If you direct an AI system, you are responsible for what you direct it to do. "The model did it" cannot become a universal excuse.

Why I came to this as a builder

I did not arrive at this from a philosophy seminar. I arrived at it from shipping. When you build in public, the way I do, you watch real people hand real parts of their lives to systems you wrote. You feel the weight of that quickly. I also studied the formal side: I completed UNESCO's Global MOOC on the Ethics of AI, because I wanted my instincts checked against people who think about this full time.

The conclusion held up. Ethics frameworks tell us what to value. Product guardrails tell a system what to block. A constitution is the missing middle layer: the standing public commitment that connects the values to the code, and survives both.

The work ahead

Nobody should write this document alone in a room. A constitution gets its authority from being shared, argued over, and adopted in the open. Builders should draft it. Users should challenge it. And every serious AI founder should be pressed to build under principles they are willing to state publicly and be judged against. I intend to.

Guardrails are where responsibility starts. A constitution is where it becomes permanent.

If your conference or audience is working through these questions, I speak on AI coexistence and constitutional guardrails. Book me for a keynote here.

I maintain the founder’s perspective on this work on the AI Coexistence Constitution page, and the canonical, versioned document itself lives at Resonatia.

← All writing Book Milan →
The Newsletter

The frontier, in your inbox.

One email when a new essay drops — the Kardashev roadmap, AI-native building, and coexistence, written as it happens. No noise, unsubscribe anytime.

You're on the list.