What is delegated authority?

Delegated authority is the power to act or decide that its holder passes to another party, within stated limits. It is usually recorded in a delegation of authority matrix setting who may approve what, up to which amount. An AI agent acting for a business exercises delegated authority, and it can never hold more than the person who delegated it.

Why does delegated authority matter once agents act?

Every action an agent takes on a business's behalf, a refund, a payment, a letter, rests on authority that started with a person. If nobody can trace that chain, the agent is acting on authority nobody gave it.

Delegation to agents also tests an old rule: you cannot delegate what you do not hold. A team leader with authority to approve refunds up to 500 cannot give an agent authority up to 1,000. A business that configures agents in a technical console, separate from its authority matrix, can break that rule without noticing. Linking the two is a theme of agent governance is a bound, not a model and of boards should govern agent authority, not agent technology.

How does delegated authority work in practice?

A delegation to an agent should state:

• Who delegates: the named person whose authority it is.
• What is delegated: the specific capability, such as issuing refunds.
• Limits: amount, counterparties, volumes, hours.
• Conditions: for example, only while the agent's certificate is in date.
• What is not delegated: decisions that stay with the delegator.
• Expiry and review: when the delegation lapses or is reviewed.

The agent's customer ceiling and limits should sit inside the delegator's own authority. A layout for recording this is the delegation of authority matrix for AI agents.

What does that look like in an example?

A housing association's board delegates authority for rent arrears arrangements to the head of income, up to a set value. She delegates to her team leaders up to half that value. When an arrears agent is introduced, she sets its limit for drafting arrangements at the team leaders' level and its authority to agree arrangements without review at a much lower figure. Every arrangement the agent agrees can be traced: agent limit, within team leader authority, within the head of income's authority, within the board's delegation. When the head of income leaves, her delegation is reviewed, and the agent's limits are reviewed with it.

How is delegated authority different from system access?

System access is a technical permission: the agent's credentials let it call the payments function. Delegated authority is a business permission: the agent may make this payment, up to this amount, on this person's authority. Access without authority is common and dangerous, because the system will execute whatever the credentials allow. The right design gives access only for what authority covers, and enforces the authority at the moment of action.

Delegation is also different from consent given by a customer. When an agent acts for a customer, for example to make a payment from their account, the authority comes from the customer's consent, not from the business. That raises separate questions, covered in PSD2 and AI agents acting for the payer: consent and authority.

Can an agent delegate to another agent?

An orchestrator at depth 5 may assign tasks to other agents, but it cannot pass on authority it does not hold, and each agent keeps its own bounds. The risk is a chain where authority becomes impossible to trace. Keeping orchestration from spreading authority further than intended is the subject of orchestration without authority sprawl.