An instruction worth keeping
Imagine asking an assistant to keep an ordinary part of life in good order: maintain the supplies you need, renew an appropriate service, or keep a small organization’s source register current. You define the purpose, the budget and the exceptions. You do not want to reconstruct the task every morning.
The appealing product is continuity. Its value comes from knowing when yesterday’s instruction still applies, when the environment has changed and when the decision belongs back with you. Twenty years is a design horizon for that relationship, not an observed product lifetime or a forecast.
From tasks to desired states
A task ends when an action is completed. A mandate persists while an agreed condition needs maintaining. “Buy these items” and “keep suitable supplies available within this budget” create different obligations. The latter requires awareness of stock, changing preferences, prices, delivery and exceptions.
The unit of value changes with it. The useful question is how much satisfactory work is maintained after fees, supervision, correction and switching costs. Interaction count is a poor substitute. A system that asks you twenty times whether to proceed may be producing activity while returning the coordination burden to you.
Silence has two meanings
A quiet assistant could be doing a very good job. It could also be doing nothing, repeating a stale policy or failing to notice that a supplier disappeared. Low interruption counts do not distinguish these explanations.
A credible service needs independent observations of the desired state. It should preserve missed cases, abandoned attempts and expensive recoveries. The customer should be able to see what happened without performing the entire job again. Quietness becomes valuable only after the outcome is established.
Durable purpose, bounded authority
The purpose of an instruction can last longer than any particular model, credential or commercial provider. Its executable authority should remain bounded. A standing objective is not permission for an agent to expand its own budget or invent a new responsibility.
The Agent Payments Protocol offers a concrete authorization design to inspect. It does not establish that a purchase was the best choice or that a service can run reliably for decades. Permission, judgment and accountability remain separate questions. [1]
The economics of not returning
A subscription can fund reliable availability. A transaction fee can fund useful execution. An outcome fee can attach payment to an agreed result. Each also creates a different tension. Transaction revenue can encourage more buying; a subscription can persist through inattention; an outcome definition can exclude the hardest failures.
Evaluate the whole service: model and tool costs, human exceptions, support, recovery and the customer’s burden. A cheap average inference cost does not settle the economics of a costly tail of failures. A profitable intermediary can serve a customer well, but its contribution has to survive this accounting.
Exit is part of the product
A lasting relationship must be able to change. The customer needs to revise the purpose, pause execution and move useful records elsewhere. A provider that retains the mandate only because departure is difficult has demonstrated persistence, not necessarily value.
The sharp product question is whether a person would choose the service again with a clear account of what it achieved and a practical opportunity to leave. Control is not the opposite of delegation. Usable control is one of the things that makes delegation possible.
What would change our mind
The thesis weakens if fixed rules provide equally good outcomes at lower total burden, if independent checking erases the attention savings, or if customers actively prefer making the supposedly repetitive decisions. It also weakens when failures make authority and exit too costly to manage.
The next useful evidence is a matched, bounded comparison with a preserved baseline. This essay proposes that test; it does not report an empirical pilot. The ambition is a service that returns time to life while keeping the person in control.
Sources & scope
Sources checked: 13 November 2026.
Authorization design; not evidence of recommendation quality or long-term reliability.