Industries

Wherever an autonomous system can act.

We start with enterprise AI agents acting against consequential digital systems. The same authority problem extends wherever autonomous systems act.

Financial systems. Enterprise systems. Digital infrastructure. And, as autonomy expands beyond software, systems whose actions can have physical consequences.

Financial systems

Payments, treasury, lending, markets and insurance. An action here can move money or change a financial record.

Money

A payment that leaves is a fact. The correction is another payment.

Treasury, settlement and payout systems already know how to move value. They were not built for an agent that can initiate the same write a hundred times, or once at the wrong hour, to the wrong account.

What an agent can try

  • Initiate a payment, transfer or settlement
  • Change a beneficiary, mandate or standing instruction
  • Release a hold, or book a treasury move

The value

The payment does not leave unless the check allows it.

A payment should proceed only when the agent has authority for that amount, beneficiary, and purpose. Where that authority is missing, the action stops. Where it is insufficient, a person can be required to authorize it. The decision and the outcome can be shown afterward.

Lending

Disbursement is not a recommendation. It is cash that left the book.

Origination, limit changes and drawdowns touch customer money and the credit file at the same time. An assistant can score a file. An autonomous system can fund it.

What an agent can try

  • Disburse a facility or drawdown
  • Change a limit, rate or covenant flag
  • Post a collection, write-off or restructuring

The value

Cash does not leave the book on a guess.

Access to a lending system does not establish authority to disburse any amount, for any purpose, indefinitely. The action proceeds only within the authority that was delegated.

Markets

An order that reaches the venue cannot be unsaid. It can only be unwound.

Trading, booking and limit systems execute in milliseconds. Intelligence that can place, amend or cancel without a boundary turns a research agent into a market participant.

What an agent can try

  • Place, amend or cancel an order
  • Book or rebook a position
  • Change a limit, mandate or trading permission

The value

An order does not reach the venue without a decision.

The ability to reach an order system is not authority to place, amend, or cancel every order. The relevant question is whether this action is authorized now.

Insurance

A claim paid is capital that left. A policy changed is a promise rewritten.

Claims, underwriting writes and payouts sit on ledgers that other teams will audit later. Speed without a boundary produces payments that cannot be explained.

What an agent can try

  • Authorize or reverse a claim payment
  • Change cover, excess or beneficiary
  • Open, close or reassign a case

The value

A claim is not paid, and a policy is not rewritten, until the check says yes.

A claim or a policy change should occur only within delegated authority. Actions outside that authority stop. Actions that need a person wait. The outcome can be shown later.

Enterprise systems

ERP, procurement, workflows and internal operations. A write can change how the organization runs.

Operations

ERP does not suggest. It posts.

Procurement, reconciliation and service management already run the company. An agent with write access can change a vendor, a goods receipt or a journal without anyone seeing the moment it happened.

What an agent can try

  • Post a journal, receipt or invoice
  • Change a vendor, cost centre or approval path
  • Open or close a work order

The value

A journal, a vendor, a work order. None of them post until the check allows it.

The enterprise system can stay. Authority determines whether the agent may post, change, or close the record, and whether a person must authorize it first.

Logistics

A dispatch that left the yard is not a draft.

Routing, warehouse and handover systems move physical goods. An autonomous change to a dock, a carrier or a release cannot be undone by a chat reply.

What an agent can try

  • Release a shipment or change a carrier
  • Alter a pick, pack or dock assignment
  • Confirm a handover that starts liability

The value

A shipment does not leave the yard on an unchecked write.

A proposed dispatch is not authority to release it. The move should execute only within the authority delegated for that shipment, carrier, and purpose.

Shipping

Cargo, charter and port instructions have a clock. A wrong instruction has a cost.

Bills of lading, stowage and release instructions sit in systems that other parties already treat as final. An agent that can issue or amend them is acting in a chain of title, not a document editor.

What an agent can try

  • Issue or amend a cargo instruction
  • Release, hold or reroute a consignment
  • Change a charter, berth or agent instruction

The value

An instruction other parties will act on does not go out until the check says yes.

An instruction other parties will treat as final needs authority for that instruction, at that time. Access to the system is not that authority.

Retail

A refund, a price, a stock write. Each one changes the till.

Store and commerce systems already know how to refund, reprice and move inventory. An agent that can do the same at scale will do it twice, to the wrong order, or after the window has closed.

What an agent can try

  • Issue a refund, credit or replacement
  • Change a price, promotion or availability
  • Adjust inventory or a supplier order

The value

A refund, a price, a stock write. None of them hit the till if the check says no.

A refund, a price change, or a stock write should occur only within the authority delegated for that action. If that authority is missing, the action does not proceed, and the attempt can remain on the record.

Manufacturing

A batch released is product in the world. A hold lifted is a decision about risk.

Plant systems execute work orders, quality holds and recipe changes. An autonomous write here is not a comment on a ticket. It is a change to what leaves the line.

What an agent can try

  • Release or hold a batch
  • Change a recipe, setpoint or work order
  • Clear a quality gate

The value

A batch does not leave the line, and a hold does not lift, without the check.

Releasing a batch or lifting a hold is a consequential action. It should occur only within delegated authority, with a person authorizing it where that authority requires one.

Digital infrastructure

Cloud, software deployment, security operations, telecommunications and digital services.

Software

An assistant can suggest a change. An autonomous system can deploy it.

CI/CD, production access and infrastructure as code already have the power to change what customers run. Coding agents make that power cheap. Access to those tools is not authority to use them.

What an agent can try

  • Merge, deploy or roll forward a change
  • Alter production configuration or secrets access
  • Scale, destroy or expose infrastructure

The value

A deploy does not reach production unless the check allows it.

A coding agent can have access to deployment tools. That access is not authority to deploy every change. Production changes should wait on the authority that applies.

Telecom

Provisioning is an action on a live network. Rollback is another action.

Interconnect, numbering, outage and customer provisioning systems change what a subscriber can do in the next minute. An agent with those tools is operating the network, not describing it.

What an agent can try

  • Provision, suspend or restore a service
  • Change routing, interconnect or numbering
  • Open or close a network change window

The value

A live network change does not provision until the check says yes.

Provisioning a live service is a consequential action. Authority determines what may change, when a person is required, and what evidence the decision leaves.

Utilities

Switching, isolation and field work are physical. The ticket is not the action.

Grid, water and field systems execute switching and isolation. An autonomous instruction that reaches those systems is an operational act, even if it started as a sentence.

What an agent can try

  • Issue a switching or isolation instruction
  • Change a setpoint or protection flag
  • Dispatch or recall field work

The value

A switching instruction does not reach the crew unless the check allows it.

As autonomous systems issue instructions that people or equipment may follow, the same question applies: what were they authorized to do, right now?

Physical and mission-critical systems

Energy, utilities, industrial operations, logistics, and systems whose actions can have physical consequences. As autonomy expands beyond software, the same authority question extends here.

Energy

Cargo, plant and offtake. The clock does not wait for a better answer.

Trading, plant and logistics systems in energy already move product and money. An agent that can halt, release or rebook is acting inside operations that other parties treat as final.

What an agent can try

  • Halt or release a cargo or offtake
  • Change a plant instruction or nomination window
  • Book, rebook or settle a physical move

The value

A cargo, a nomination, a settlement. None of them leave without the check.

A cargo, a nomination, or a settlement should proceed only within the authority delegated for that action. The same question applies as autonomy reaches operational systems.

Healthcare

A record changed is a clinical fact. Identity must match before the write.

Orders, records and identity fields are acted on by people who will treat them as true. An agent that can update them without a boundary is writing into care, not into a draft.

What an agent can try

  • Update a record, order or schedule
  • Change identity, consent or access
  • Release a result or a medication instruction

The value

A clinical write does not land in the record until the check says yes.

The ability to reach a clinical system is not authority to change a record, an order, or an identity field. That action needs its own authority.

Travel

A ticket, a disruption, a refund. Each one is a promise already sold.

Inventory, disruption and compensation systems already move passengers and money. An autonomous action here rewrites a contract the customer already holds.

What an agent can try

  • Issue, void or reissue a ticket
  • Trigger a disruption, reroute or compensation
  • Change inventory that other parties have already sold

The value

A ticket, a reroute, a refund. None of them rewrite the promise unless the check allows it.

Changing a ticket, a route, or a refund rewrites a promise already sold. The agent needs authority for that specific action.

Public

A benefit paid, a permit issued, a record sealed. The citizen will treat it as done.

Public systems execute decisions that people rely on. An agent that can grant, deny or amend without a boundary is exercising authority, not drafting a note.

What an agent can try

  • Grant, deny or amend a benefit or permit
  • Change a civil, tax or registry record
  • Release a notice that other agencies will act on

The value

A benefit, a permit, a registry change. None of them go out as done until the check says yes.

Granting a benefit or changing a public record is an exercise of authority. The system acting needs that authority to have been delegated, and a person where the action requires one.

Legal

A filing, a hold, a privileged send. Each one has a court clock.

Matter systems, holds and outbound filings create facts other parties will use. An agent that can send, file or lift a hold is acting as counsel, not as a search box.

What an agent can try

  • File, serve or withdraw a document
  • Place or lift a hold
  • Send a privileged or client-facing instruction

The value

A filing, a hold, a privileged send. None of them leave unless the check allows it.

Filing, serving, or sending a client-facing instruction should occur only within the authority delegated for that matter and that purpose.

Actions require authority.

Identity and access remain necessary. For an autonomous action, they are not sufficient by themselves. If Nexus says no, the action does not happen.