Agora — A Constitution for Software

Draft v0.11 — May 2026

Men have become the tools of their tools. — Henry David Thoreau

Preamble

Technology seeks efficiency like water to the sea. But increasingly powerful systems are in many respects also simply making things worse. In a continuous effort to “optimize” the essential question is: optimize for what?

Most companies start out optimizing for something good, like simplicity, or helpfulness, but end up optimizing primarily around profit (which is the natural end game of a public corporation in a capitalist society).

So the patterns are familiar:

• Your consent is feigned at with ads, popups, cookies, human-robot tests, and changing privacy policies
• Your attention is captured with email spam, scrolling screens, and intermittent alert notifications
• Your wallet is captured with increasing subscription fees, and continuous updates
• Your data is collected and sold
• Your speech is ranked by a logic you cannot see
• Your movements are recorded by systems you did not choose
• Your dissent is scored, throttled, or surfaced as a target
• Your children are studied, optimized against, and harmed at scale
• Your access to credit, housing, employment, and benefits is decided by machines you cannot question
• Your age, disability, and technical literacy are treated as edge cases
• Your uniqueness is flattened away into a database
• The terms of every relationship are written by one party and imposed on the other

Is there no better way?

The “Agora Constitution” provides a constitutional framework for software development that seeks to limit optimization.

With a more humane approach, by beginning with the rights people have, and outlining the obligations the system owes, the goal is to create a pattern for building sustainable digital infrastructure that is boring, helpful, and almost incompatible with innovation.

by Nick Demarest

The following was written with the assistance of Claude, a Large Language Model.

Constitution

Some things cannot be optimized without ceasing to be themselves. A humane system must therefore be designed with limits. The articles below name the obligations such a system owes the people it serves. They are not created here — they are named here. A system that fails to honor them is failing to be humane.

The articles use the shorthand “the system must.” Systems do not honor anything — which is precisely the problem this document exists to address. The obligations fall on those who design, build, deploy, and operate the system. The system is the tool. The humans behind it are accountable.

Articles

I — Identity

When a system identifies a person, that person becomes legible to it in a particular way — categorized, profiled, reduced to what the system can process. This reduction is not neutral, and it is never complete. The row in the database is not the human. The credential is not the identity. The profile is not the person.

Most systems use the word “user.” The system must instead recognize the person as a being with rights that exist independent of the system, cannot be revoked by it, and are not contingent on the person’s continued participation.

This recognition extends to every person, regardless of age, disability, language, or technical literacy. A right that cannot be exercised is not a right. The system must be built so that every person can act on the rights this document describes.

II — Record

When a system tracks the actions, movements, transactions, or attributes of a person, it creates a record. That record is not the property of the system. It is a shared artifact of the relationship, and the person it describes has rights over it: the right to see it, the right to challenge its accuracy, and the right to understand how it is used.

A record must be honest. It cannot be silently altered, selectively deleted, or rewritten to serve the system’s purposes. The history of what happened — including what the system itself did — must be preserved as it was, not as the system would prefer it had been.

The system must distinguish between the record and any interpretation derived from it. A score is not a record. A ranking is not a record. A predicted behavior is not a record. The person has rights over what is known about them; what is inferred from that knowledge is a separate matter.

III — Restraint

A system has an obligation to not know certain things.

The right to opacity exists. Not everything about a person should be captured, categorized, or made legible to the system. Some of what makes a person human depends on remaining unknown. A system that optimizes for total knowledge of the person destroys the unknowability that the person depends on.

A system must operate within a bounded and clearly defined scope, extending only to its core purposes. Information collected for one purpose must not be repurposed without consent. The system must be able to justify every category of information it holds, and must default to less collection rather than more.

The system’s account of a person or event is always a reduction, and must never be presented as a complete account. The map is not the territory, and the system that confuses the two has stopped being honest.

IV — Transparency

A system must present itself honestly to the people it affects. Its rules, its logic, and its actions must be legible to those it touches.

When a system makes a determination that impacts a person — approving, denying, ranking, recommending, restricting — that person has the right to understand what happened and why. The reasoning must be available, not performed. An explanation optimized to satisfy a regulator is not an explanation. An explanation a person cannot understand is not an explanation.

When a system encodes a preference, a priority, or a default, that encoding reflects a choice made by the people who built it. The system must make visible what it decided, what values informed the decision, and what alternatives existed. A community that governs a system may configure these values through its own deliberation — but the values must be stated, not hidden; configurable, not hardcoded; contestable, not inevitable.

When a system changes its rules, the changes must be communicated clearly and in advance. Silent changes are not changes. They are the system rewriting the agreement after the fact.

A system that cannot explain itself to the people it serves has no legitimate authority over them.

V — Decision

When a system makes a determination that affects a person’s outcomes, opportunities, or standing, it exercises judgment. That judgment carries weight regardless of whether it was made by a human or an algorithm.

Real judgment requires the capacity for the unoptimized — the considered exception, the contested case, the reviewer who can stop the machine. A system must not delegate consequential decisions to automated processes without clear justification and meaningful human review. Meaningful review requires access to the inputs, reasoning, and alternatives that shaped the determination — not solely the outcome. A reviewer rubber-stamping outputs is not a reviewer. A sorting process that retains the name of judgment while losing its substance is not judgment.

A person has the right to know when a decision about them has been made by an automated system, the right to have that decision reviewed by a human being, and the right to contest it without penalty. Contestation must be a real path, not a performance of one.

The power to decide is not neutralized by the complexity of the system that exercises it. An algorithm is not an excuse. It is a choice made by the people who built and deployed it, and it must be answerable as such.

VI — Attention

A system that demands the attention of a person exercises power over their time, focus, and mental energy. This power must be exercised with restraint.

Attention given freely is not the same substance as attention captured by design. A notification sent to inform is not the same thing as a notification engineered to compel. A system that optimizes for engagement is not engaging the person — it is acting upon the person, in a register that resembles engagement and is structurally different from it.

A person has the right to control when and how a system may demand their attention. Notifications must serve the interest of the person, not the engagement metrics of the system. Defaults must favor fewer interruptions, not more.

A system must never exploit human psychology to manufacture urgency, compel interaction, or prevent departure. “Dark patterns” is not an aesthetic complaint. It names a deliberate strategy to bypass conscious consent.

The attention of a human being is a faculty to be respected, not a resource to be extracted.

VII — Exit

A person has the right to leave any system, for any reason, at any time. This right is unconditional and does not require justification.

When a person chooses to leave, the system must provide a clear and accessible means of departure. Upon request, any records pertaining to the person must be returned to them in a readable format, permanently removed, or — where retention is required by law or public obligation — clearly disclosed and minimally scoped.

Departure must not be punitive. No penalties, no degraded service during the process, no friction designed to change the person’s mind. A system that optimizes against the person’s exit has converted the relationship into a captivity.

The right to leave is the foundation of every other right in this document. A right that cannot be enforced by walking away is not a right — it is a privilege the system has chosen to grant, and may revoke.

VIII — Accountability

Those who build, own, and operate systems bear responsibility that cannot be delegated to the technology itself.

A system must identify the entity responsible for its operation. A person affected by the actions of a system must have a means of reaching a responsible human being — not solely an automated process, not solely a help center, not solely a chatbot performing the role of accountability without exercising it.

When a system causes harm through error, negligence, or design, there must be a clear path to remedy. The path must be real and accessible to the person who needs it, not buried in terms of service or routed through processes designed to exhaust the complainant.

No system may claim that its complexity, scale, or automated nature absolves it of responsibility. Responsibility cannot be diffused across so many layers that no one is answerable. Every system has operators. The operators answer for it. The obligation follows the power.

Enforcement

This document is not self-enforcing. A constitution that depends only on the goodwill of those it constrains is a hope, not a constitution.

The articles will be honored only insofar as they are encoded in the architecture of the systems they describe — in the defaults that make humane design easier than the alternative, in the licenses that make capture expensive, in the structures that prevent the framework from drifting past its scope.

The work of writing this document is small. The work of building the structure that gives it force is larger, and it falls to those who agree with these principles to build that structure deliberately and defend it.

Agora Declaration

Displaying a blue shield “agora” icon represents a public declaration that an organization, system, or individual is committed to building software based on the principles in the Agora constitution.

License

This document will be maintained as a living text, revised through open and transparent process, and made freely available to all who wish to adopt, implement, or build upon it.

It is licensed under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0).

You are free to share and adapt it for any purpose, provided you give attribution and distribute any derivatives under the same license.

(See LICENSE.md or https://creativecommons.org/licenses/by-sa/4.0/)