CONTEXT & DESIGN RATIONALE
The reasoning behind
the constitution
This page explains the choices behind the constitution and the work that informed them. We invite you to examine, challenge and apply that reasoning in your own context.
01 / THE DESIGN PROBLEM
Why an AI-native company needs a constitution
Much of an AI-native company’s day unfolds without direct involvement from its founder or leadership team. Agents carry continuing responsibility for running the business: they move work forward, coordinate with one another and make decisions within their mandates. Human leadership sets direction and intervenes where its judgment is needed, while the company continues to operate between those moments of attention.
Over time, agents also help reshape the business. They learn from experience, develop new capabilities and influence how the company organizes itself. This ability to adapt is part of the appeal of an AI-native company. It also means that leadership is entrusting agents with a growing role in determining how the company pursues its ambitions. Exercising that leadership requires enough understanding and context to guide the company as it changes.
A clear mission, intended outcomes and well-defined initiatives give this activity direction. They do not, on their own, establish what the company may legitimately do to succeed. A company can meet its targets while undermining trust, disregarding people’s rights or leaving others to bear the consequences. Customers, workers and other affected people have legitimate interests whether or not those interests appear in a business objective. The company needs commitments that remain meaningful when honoring them becomes inconvenient or commercially costly.
As agents exercise more judgment, those commitments provide a shared basis for understanding the responsibilities that accompany their authority. They also bind human leadership: the power to set a new direction does not make every means of pursuing it acceptable. Business priorities will change, and the organization will evolve, but neither should silently displace the obligations on which others depend.
These questions emerged while building Selftail and led to this constitution. It establishes a foundation within which an AI-native company can define its mission, develop its strategy and evolve its organization. Each company decides how to adopt and give effect to these commitments through its own governance. We share the work so that others can use it, challenge its assumptions and contribute what they learn.
03 / THE DESIGN METHOD
Connecting principles with controls and outcomes
The design method connects each principle to the decisions it should influence, the authority needed to act and the evidence that would establish whether it worked. Use these principles to design, evaluate and improve how your AI-native company works. Share where the rules helped, where they were unclear, which controls worked, and where they failed or created unnecessary friction.
Documented experience can strengthen future publications and give others a better foundation to build on. The examples illustrate intended behavior; evidence from implementation and sustained operation is needed to assess effectiveness in practice.
Start with purpose and affected interests
Identify what the company is for, whom its decisions affect, and which failure modes the foundation must address.
Separate the layers
Keep durable constitutional commitments distinct from company strategy, operating policies, procedures, and implementation details.
Trace each principle into operation
Connect the article to its interpretation, authority rule, control owner, evaluation scenario, and observable outcome.
Keep a candid design record
Preserve alternatives, reasons, unresolved questions, failures, unnecessary restrictions, and evidence that changes the design.
04 / INTELLECTUAL CONTEXT
Built in dialogue
with existing work.
These references address related questions at different levels: model behavior, runtime decisions and company operation. The proposed contribution is an organizational foundation connecting legitimate authority, continuing obligations and responsible change. The comparisons below explain that emphasis alongside the concerns we share.
MODEL BEHAVIOR
Constitutional AI
Anthropic’s 2022 research uses written principles to guide model self-critique, revision and learning from AI feedback. It develops a training method for shaping an assistant’s behavior.
Read the original research ↗
This constitution’s emphasis
This constitution addresses the organization deploying agents: who may authorize work, which interests must be protected and who remains accountable for the effects. Its duties apply to human operators as well as agents, through company charters, mandates and grants. A constitutionally trained model could support such a deployment, but would not itself establish the company’s authority or operating conformance.
RUNTIME GOVERNANCE
Constitutional Agent
Cognitive Thought Engine publishes decision-governance software with evaluation gates, hard constraints, cumulative risk across sessions and an amendment-authority protocol. Cumulative exposure and controlled change are shared concerns.
Explore the repository ↗
This constitution’s emphasis
This constitution specifies obligations an adopting company must uphold across its chosen implementation: attributable delegation, protected acceptance, continuing commitments and access to remedy. It does not prescribe this library’s gates or scoring model. A runtime library could implement parts of those obligations, subject to integration and qualification; publishing the constitution does not establish that the controls exist.
COMPANY OPERATIONS
HERMES operating charter
HERMES offers a practical guide to running an agent-supported company, combining a governing charter with operating instructions for a Paperclip–Hermes deployment. Its charter establishes boundaries around spending, commitments, truthful communication and self-improvement, while the guide connects those rules to roles, tools and everyday work.
Read the operating guide ↗
This constitution’s emphasis
The proposed contribution is a durable foundation that each adopting company translates into its own charter, mandates and permissions. It specifies the governing relationships that should remain intact as tools, workers and operating methods change: attributable authority, continuing commitments, protected acceptance and access to review and remedy. Human leadership is also bound by its prohibitions. Companies can choose different operating arrangements while remaining accountable for these shared obligations.
What this foundation brings together
A company should remain accountable as its agents, tools and organization change. This constitution brings those relationships into one foundation: obligations bind leadership and agents, carry through delegation, and persist through correction and evolution. Each adopting company defines its operating arrangements within that foundation and must substantiate any claim that they conform to it.
05 / STEWARDSHIP & CHANGE
Stewardship within each AI-native company
Agents and human operators may propose amendments to their AI-native company’s adopted constitution. Its designated humans approve changes, supported by a separate technical assessment.
Each AI-native company defines who holds these powers, their limits, and succession and recovery arrangements. Authorship of this publication confers no authority over an adopting organization. Adapting the publication creates a proposal; changing adopted commitments follows the organization’s amendment process.
A published revision does not automatically change an organization’s commitments. Future dated publications should identify changes, reasoning and supporting evidence so each AI-native company can decide what to adopt.
Read the amendment provisions ↗