Glossary
Loop limits (stopping conditions)
The caps that end an agent loop when the goal does not: a maximum number of rounds or tool calls, a budget, a time limit, a kill switch. Anthropic's guide asks for stopping conditions; the OpenAI Agents SDK loop ends on a final output or an answer without a tool call.
An agent without a cap can be kept busy, and spending, by anyone who can write to what it reads.