Triage AI cuts incident MTTR with Quick and New Relic

One prompt gives instant triage as AI handles incident response and RCA.

Incident triage is a high-stakes, time sunk cost in engineering teams: gathering evidence, assessing user impact, and handing off work across tools can drag on long enough to lose context between shifts. The latest pattern from AWS shows how a custom incident triage assistant can orchestrate those steps in a single conversational flow. Built with Amazon Quick and New Relic, the agent coordinates with the New Relic MCP Server and Asana through native integrations, so a single prompt can trigger evidence gathering, root cause analysis, and a tracked handoff task. The team reports that this approach reduces the evidence-gathering phase and speeds resolution, while preserving a consistent investigation standard across on-call rotations.

At the heart of the setup is Amazon Quick, a family of chat agents designed to connect enterprise tools through pre-built integrations called connectors. The New Relic integration is built in as a native connector, enabling Quick to reason about telemetry and incident data and then push outcomes into actionable steps. The workflow is orchestrated with the New Relic MCP Server protocol, which provides a structured, model-backed context for the agent to pull in relevant evidence links, logs, traces, and performance signals. When the agent finishes its triage per the prompt, it creates an Asana task that’s ready for the on-call handoff, ensuring nothing falls through the cracks in shift changes. In practice, this turns a multi-tool investigation into a single, auditable thread that includes a concise RCA brief and direct next steps.

The engineering value is clear: triage that previously required scrolling through dashboards and switching between Jira, Slack, email, and tickets can now be captured in one coherent narrative. In internal testing against New Relic’s own applications, the approach reduced the evidence-gathering portion of triage, shortened time-to-resolution, and standardized the investigative approach across teams. The broader message is that tools can be choreographed so that engineers spend less time assembling a story and more time correcting the issue, with the handoff documented in a centralized task system.

From a practitioner’s standpoint, several constraints shape how well this pattern lands. Data freshness and connector reliability matter: if telemetry arrives late or a connector misreads an event, the RCA brief risks being incomplete or off-target. Latency in cross-tool orchestration is another hidden cost; while a single prompt is powerful, the underlying data fetches and task updates can introduce small delays that matter in high-pressure incidents. A failure mode to watch for is premature or overconfident auto-generated RCAs; teams should enforce human validation for critical incidents and ensure evidence links are preserved in a retrievable form. Governance and access control also matter: queuing sensitive incident data behind authentication and maintaining an auditable trail across tools protects against leakage and misattribution. Finally, expansion potential is real but non-trivial: adding more connectors (for Jira, Slack, or security tooling) can broaden coverage, but requires careful maintenance of data schemas and consent flows to keep investigations reliable and reproducible.

Looking ahead, the pattern aligns with a broader capability in Amazon Quick: connecting enterprise tools to AI agents through native integrations. The combination of Quick’s orchestration, MCP Server reasoning, and targeted tooling like Asana for handoffs already offers a template for scalable, teachable incident workflows. The challenge for teams will be balancing the speed and accuracy of automated triage with guardrails that preserve reliability, privacy, and the human-in-the-loop where it matters most.