Agentic Monitoring
2026-10-31
![]()
Benjamin Congdon describes using AI agents as monitors, the way Anthropic does: "superfans" of a system, with access to its telemetry and logs, who notice the subtle anomalies that static alerts miss, such as a change point just after a production deploy. Since most outages follow a change, about 70% by his count, that is where they earn their keep. He argues it is a step up from expensive observability platforms that needs no new SaaS product, only current models pointed at data you already have.
Was this useful?