Agent run history & budgets
Automation needs visible execution history because an agent or workflow can make multiple model and tool calls without a person manually sending every step.Run history
A useful run record can include:- run ID
- owner/project/organization
- agent
- start and end time
- status
- model usage
- tool calls
- credits
- approval waits
- result or artifact references
- failure stage
Budgets
Runtime budgets can limit:- total credits
- model calls
- tool calls
- duration
- concurrent runs
- allowed models
- allowed tools
Budget exhaustion
When a hard run budget is exhausted, the workflow should stop or enter a defined failure/waiting state rather than continuing indefinitely. The run history should distinguish budget exhaustion from a provider or logic failure.Investigate expensive runs
Look for:- repeated repair/retry loops
- unnecessarily deep reasoning
- broad search/retrieval queries
- excessive tool calls
- repeated failed steps
- large context injected on every iteration
Retry carefully
Do not retry a failed run before understanding whether it failed because of a temporary service issue, an invalid workflow definition, or an exhausted budget.Organization attribution
Runs executed in an organization context should remain attributable to the user/application and workload that caused the usage even when credits come from a shared pool.Credits & metering
Understand how automation usage contributes to billing and credit accounting.