Agent Loop Budget canary
Agent Loop Budget canary — AI retrieval, agent, evaluation, or production-reliability concept.
Definition
Agent Loop Budget is the hard cap on how far an AI agent may run its plan→act→observe loop: typically max steps, tokens, tool calls, wall-clock time, and dollar cost. Agent Loop Budget canary is that budget object applied in a canary context — the agent must stop or degrade gracefully when any limit is hit.
Why it matters
Without a loop budget, tool-using agents can recurse, re-retrieve, and burn spend while looking “busy.” Budgets turn agent autonomy into an operable production control: measurable, alertable, and comparable across canary/shadow/production.
Case
A canary rollout lets 5% of research queries use a new tool-calling agent while production stays on the old path. If token burn or tool-error rate spikes, you kill the canary before the whole desk inherits the failure.
How to read it
Read the numeric limits (steps/tokens/tools/time/cost), the stop behavior, and which environment owns the budget. A canary budget that never trips is untested; a production budget that always trips means the workflow is mis-sized.
Ask the macro AI about this object
Opens ZChat with Codex, RAG, and chart context enabled. Connected to the shared Ztrader memory layer.