How to Define SLOs for AI Agents Beyond Uptime

An agent endpoint returns HTTP 200 in 99.95% of requests. The service is “up.” Meanwhile, some successful responses skip retrieval, some tool calls exceed their budget, and a few write actions occur w