AI and Intelligent Automation
AI and Intelligent Automation
This domain tests how New Relic applies AI to observability: AI-assisted querying and analysis, monitoring of LLM-backed applications, and automated detection and routing of incidents.
Core rules
- The New Relic AI assistant in the UI generates NRQL, explains charts, and summarizes incidents. Its output is a starting point; validate attribute names, event types, and time ranges before trusting it.
- AI monitoring instruments LLM applications: OpenAI, Azure OpenAI, AWS Bedrock, LangChain, and LlamaIndex integrations emit traces and spans recording prompts, completions, token counts, latency, and model name, plus feedback and evaluation events for quality scoring.
- Applied Intelligence was the legacy AIOps product; its functions now live in alerting, incident intelligence, and workloads. Do not describe it as the current product.
- Alert conditions support static thresholds and anomaly/baseline conditions that learn normal behavior per signal and time of day.
- Automation routes incidents through workflows to destinations such as email, Slack, PagerDuty, ServiceNow, or a webhook; muting rules suppress noise during maintenance.
- Incident correlation groups related failures so responders see one incident instead of dozens.
Worked example
A checkout service's traffic triples every weekday at 09:00. A static 400 ms latency threshold fires every morning although nothing is wrong. Switching that condition to an anomaly/baseline condition, which compares current behavior to the learned pattern for that time window, keeps the alert for genuine deviations, while the static rule stays on error rate where a fixed limit is meaningful.
Traps
- Copying AI-generated NRQL without checking attribute names or the event type.
- Assuming LLM monitoring works with no instrumentation; the SDK or library must export spans to New Relic.
- Creating alert conditions with no workflow or destination, so nobody is notified.
- Confusing incident correlation (grouping) with muting (suppression).