Agent-driven SLA watcher with judgment-based escalation

When a ticket is flagged as predicted-to-breach, this agent picks it up and decides what to actually do about it. It reads the conversation, checks whether the ticket is blocked on the customer or on the team, gauges severity, and only escalates the ones that truly need it — choosing the right destination (a Linear engineering issue for a product bug, a PagerDuty page for an outage-class miss) and writing a clear rationale.

Use it when a pure timer-based forecast produces too much noise and you want human-like triage at scale. The agent filters out tickets that are legitimately waiting on the customer and reserves escalation for real, actionable risk.

1. 1A webhook fires when an upstream forecast marks a ticket as predicted-to-breach.
2. 2The agent pulls the full ticket thread and metadata for context.
3. 3It reasons over urgency, blocker, and severity to decide whether and how to escalate.
4. 4For product defects it opens a Linear issue with reproduction context.
5. 5For outage-class or contractual misses it raises a PagerDuty incident.
6. 6It posts its decision and rationale to the team channel and closes the loop.