60-80% of overnight pages are routine patterns the on-call could fix with a runbook, but still wake a human up to run.
On-call burden is the single biggest driver of senior SRE attrition, and the painful truth is that most pages are routine. The pod that needs a restart, the cache that needs a flush, the queue that needs a drain. Each is a 30-second action a human could do in their sleep, except they can't, because they have to wake up first, get to a laptop, and navigate four tools to do it. The cumulative damage is measured in attrition.
Nova's policy envelope decides which pages need a human and which don't. Routine patterns close themselves; novel ones still page.
You define what Nova can do without approval (scale this deployment up to N replicas, restart this pod, flush this cache, rotate this key) and what requires a human (anything affecting customer data, anything with blast radius > 1 service).
Agents earn autonomy by track record. An action with 500 successful runs and zero rollbacks gets a trust score of 99 and operates without approval. An action with mixed history requires a human.
When the symptom doesn't match a known pattern, or the trust score is below threshold, Nova pages the on-call with full context, the alert, the correlated signals, the candidate runbooks, and the recommended action.
Get Started Free, cancel anytime. Or book a 30-minute walkthrough with a founder.