Who Owns the Agent?

The Intent Architecture Stack is a three-layer organizational design framework that specifies how to build accountability into an AI agent's governance record before it goes live. Layer 1, Context, maps the regulatory environment, affected stakeholders, and system integrations before intent is defined. Layer 2, Intent, documents the agent's authorized purpose, explicit prohibitions, and expected outputs. Layer 3, Governance, names the Consequence Owner, defines the review cadence, and establishes the escalation path. Together, the three layers produce the documentation set that answers the board's accountability question from a record that predates any incident.

Auditability means the organization can reconstruct what happened, including which agent took which action, when, and on what data. Accountability means the organization can identify, from a document written before the incident, who authorized the agent's action scope, who owns the consequence, and what organizational structure connects those two things. An organization can have complete auditability and zero accountability simultaneously. The logs show every action the agent took. The board asks who authorized those actions. Those are different questions, and only one survives a board meeting.

NIST AI RMF, the EU AI Act, ISO 42001, CSA Agentic Trust Framework, and SR 26-2 all point back to organizational governance. SR 26-2 Footnote 3 explicitly excludes generative AI and agentic AI from its model risk management scope as novel and rapidly evolving, while directing institutions to apply existing governance practices to tools and systems outside the guidance's scope. None of these frameworks specify the organizational design that makes accountability operational at the individual agent level. The Intent Architecture Stack is offered as one implementation pattern for that gap.

A Consequence Owner is the named individual who has formally accepted accountability for an AI agent's behavior before the agent enters production. They are not the developer who built the agent or the IT team that approved the integration. They are the business owner, the person who can answer "yes, this agent should exist and here is why" if a regulator or board member calls. Their name is in the authorization record. That record predates the agent's first live action. A populated sponsor field in a platform registry is not this designation. It is the platform record that points to the governance documentation. The documentation is what gives the platform record its organizational meaning.