Authority and Delegation

Agentic AI changes the meaning of permission.

In a conventional enterprise system, permission often describes access. A person can view a record, edit a document, approve an expense, execute a payment, or deploy an application.

Agentic execution introduces a more complex form of participation.

A system can interpret a situation, choose among available actions, and carry work across several enterprise boundaries. It can do this at speed, at scale, and across time. The enterprise therefore needs to express authority with precision.

Capability describes what a system can do. Delegation defines what the enterprise authorizes it to do.

This distinction sits at the center of Agentic AI governance.

The Decision to Delegate asks which work and level of participation the enterprise should authorize. This chapter specifies that choice: the operating envelope, accountable principal, escalation conditions, intervention rights, and evidence required before execution.

Authority has several dimensions

Delegated authority is more than a credential.

It includes the purpose of the work, the accountable principal, the actions available to the system, the resources it may use, the conditions that govern action, the duration of the authority, and the human owners who can intervene.

These dimensions create an operating envelope.

A service agent may resolve a delivery exception up to a defined financial threshold. A finance agent may prepare a correction package and route it through an approval path. An engineering agent may investigate an incident, assemble a remediation plan, validate the proposed change, and submit it into a controlled deployment process.

Each form of authority serves a specific outcome and carries a specific consequence.

Delegation follows the work

The strongest authority architecture begins with the workflow.

It asks what work must happen, which decisions appear inside that work, which actions carry material consequence, and where human judgment adds distinctive value.

This creates a delegation gradient.

At the first level, the system gathers information and prepares insight.

At the next level, it recommends a decision.

It can then prepare execution for an authorized owner.

It can execute bounded actions through defined conditions.

It can also coordinate a sequence of actions across systems and functions, while assurance mechanisms observe the work and route exceptions toward accountable owners.

Each level requires a clear connection between authority and accountability.

The enterprise can expand the operating envelope as evidence demonstrates quality, reliability, and economic value.

Delegated execution authority can also be narrowed, suspended or revoked when conditions change. Where authorities conflict, the operating model must establish precedence before execution.

This approach creates a path from assistance to increasingly capable execution.

Identity establishes the acting participant

Every consequential action needs a visible identity.

The enterprise needs to know which system acted, which workflow initiated the action, which delegated authority applied, which tools carried the action into external systems, and which owner remained accountable for the result.

This identity can travel through a chain of execution.

Delegation can narrow authority; it cannot silently expand it. Each consequential downstream action should be authorized against both the participant performing it and the authority delegated by the original principal, and should not exceed the narrower of the two.

A customer service agent may call an inventory service. The inventory service may initiate a fulfillment workflow. The fulfillment workflow may engage a carrier platform. Each participant carries a role in the broader action path.

The enterprise gains confidence when this chain remains visible.

Identity connects execution to ownership. Delegation connects identity to authority. Evidence connects authority to outcome.

Together, they create a traceable system of action.

Boundaries create confidence

Authority boundaries help the enterprise move faster.

They clarify which work can proceed immediately, which work needs approval, which conditions require escalation, and which owners can intervene.

A well designed boundary reduces ambiguity for both humans and systems.

Consider a customer remediation workflow.

Consider a hypothetical authority specification for this customer remediation workflow. Its purpose is an approved replacement. The service policy names the accountable principal and permits the replacement action only when inventory and the commercial threshold satisfy policy. The delegation identifies the tools and resources available, an expiry or renewal condition, and the evidence to retain. A larger concession goes to the service leader, a disputed contract term goes to the commercial owner, and a named intervention owner can pause or terminate the action path. If a required condition cannot be established, the system routes the case rather than treating missing evidence as permission.

Each action sits inside a clear boundary.

The system can carry routine work with speed. Human judgment focuses on material consequence, complex tradeoffs, relationships, and exceptions.

This is how agentic architecture creates a stronger combined operating model.

Accountability remains human

Machine execution can carry work. Human leaders retain responsibility for the purpose, policy, and consequence of that work.

Accountability sits with the leaders who define the outcome, authorize the operating envelope, own the relevant function, and govern the conditions under which the system participates.

The decision rights ladder in The AI Native Operating Model, Chapter 05 and the authority map in the Delegated Action Architecture (FW 18) address the same question at different levels.

This creates a shared responsibility model.

Technology leaders own the execution environment, reliability, security, and controls.

Functional leaders own the business outcome, workflow design, decision rights, and acceptance criteria.

Risk, legal, finance, and security leaders shape the policies and evidence required for consequential action.

Executive leadership determines the strategic purpose, investment priority, and authority model.

Agentic execution succeeds when these responsibilities compose into one operating system.

The executive question

The executive question is not:

“How much autonomy can the system have?”

The stronger question is:

“Which authority can the enterprise delegate, under which conditions, with what evidence, and who owns the consequence?”

This question gives leaders a practical path toward agentic capability.

It begins with an outcome. It defines the work. It assigns decision rights. It creates an operating envelope. It establishes evidence. It expands through demonstrated confidence.

The result is an enterprise that can place machine capability inside accountable authority.

That is the foundation of delegated action.