What the machine’s work costs is visible when it matters — estimated before the run, metered during it, and bounded by quotas the actor can see against their remaining allowance.
Forms
Recurring forms: pre-run estimates of time and money, live token or cost meters, remaining-quota displays, cheap draft modes offered before the expensive run, and hard ceilings that stop work at the limit.
Resources & references
- Shape of AI — Cost estimates, Draft mode
- aiuxpatterns.com — Generation Tokens
- AI UX Playground — Pre-Task Cost Estimate, Running Meters, Hard Budget Ceilings
To-do
The boundary with bounded autonomy (visibility here, authority there) is drawn but untested. The rationing force could borrow from metered-utility research well older than AI.
Consequences
- the cost of asking is visible before and during the work — estimates up front, meters while it runs, quotas in view — so spending becomes a decision instead of a discovery
- visible meters change behaviour: actors ration exploration, and the counter's presence taxes exactly the open-ended use the system is best at
- cost talk imports the provider's economics into the interaction; the unit shown (tokens, credits, currency) frames what the actor thinks they are buying
Related patterns
Serves
- Delegation — the run-cost leg of monitoring — what the delegated work is consuming while it runs
Enacts
- Agency — spending the actor can see is spending they actually consented to
Complements
- Bounded autonomy — a budget only binds when spend is visible against it
- Plan preview — cost estimates are the spend half of a previewed plan — what it will do and what it will cost
Related
- Living document — standing automated work is where unbudgeted cost quietly accrues
- Status feedback — a running meter is state change delivered at the cadence it matters