Welcome to New Relic Autopilot, formerly known as SRE Agent. We're excited to introduce you to a new way of managing incidents. Autopilot is designed to assist in incident management workflows by helping to automate certain data correlation and diagnosis.
Prudence
Compliance restrictions: Autopilot may be used with HIPAA accounts. Autopilot is not yet available for FedRAMP.
Retention periods for history/MELT data in the agent may differ from your general account retention settings.
New Relic Autopilot includes access to Generative AI resources and is subject to the Generative AI Service Specific Terms.
Autopilot suggestions should be used for your internal business purposes only.
About New Relic Autopilot
New Relic Autopilot is an AI-powered operations experience that helps engineering teams investigate issues, understand system behavior, and accelerate operational workflows.
How Autopilot works
Autopilot combines New Relic telemetry, AI reasoning, and operational workflows to investigate production issues. It coordinates specialized capabilities that gather context, analyze signals, identify likely root causes, and recommend next actions, while keeping engineers in control of operational decisions.
Capabilities
Here are some examples of how Autopilot can help best support your operations:
Alert triage and validation: The agent is designed to act as a preliminary filter for incoming notifications. It can help evaluate the context of an alert against historical baselines and current system health to assist in determining if an event is a high-priority anomaly or an expected fluctuation.
Real-time entity health: The agent can provide a quick, high-level pulse check on monitored parts of your stack. By attempting to correlate available golden signal metrics with underlying infrastructure performance, it aims to generate a narrative summary of system health.
Change intelligence: It can help identify potential performance regressions, new error patterns, or latency spikes introduced by recent updates to assist in checking if your system is healthy after deployments.
Deep incident analysis: When system degradation occurs, the agent is designed to assist with automated root-cause discovery. It can help traverse distributed traces, analyze log patterns, and inspect resource metrics to help identify potential technical failures or downstream dependencies that may be contributing to an outage.
Remediation suggestions: Beyond diagnosis, Autopilot aims to assist with recovery efforts. Once a potential root cause is identified, the agent may suggest potential technical interventions—such as rolling back a deployment or adjusting resource allocations—which an engineer should review before implementation.
Slack collaboration: Use Autopilot directly in Slack. Mention the New Relic app in a channel to triage alerts and investigate incidents without leaving the thread. Refer to Autopilot Slack integration.
Important
Autopilot suggestions should be reviewed carefully by you for your use case.
Operational workflow
The following workflow illustrates how Autopilot can help transform telemetry into actionable operational intelligence. Each stage builds on the previous one, allowing engineers to move from detection to diagnosis and recommended remediation more quickly.
How Autopilot analyzes your environment
Autopilot is designed to act as an intelligent layer between your telemetry data and potential insights. When you ask the agent a question, it performs an analysis by leveraging a specialized suite of tools.
The agent's analysis process follows three primary pillars:
Comprehensive context gathering: The agent is designed to gather relevant details, such as mapping available architecture. It can help identify relevant entities, explore their relationships and dependencies, and highlight recent change events or deployments that may have triggered a shift in performance.
Deep telemetry investigation: After the context is established, the agent dives into the problem. It can help evaluate:
- Metrics and trends: Analyzing throughput, latency, and error rates attempting to forecast potential future system behavior.
- Logs and errors: Sifting through distributed traces, error groups, and log patterns to help isolate potential sources and causes of an anomaly.
- Health and alerts: Reviewing active incidents, alert policies, and security vulnerabilities to understand the potential blast radius.
Automated root cause and impact analysis: The agent is designed to assist with causal analysis. It attempts to correlate performance dips with specific transaction patterns, infrastructure bottlenecks, and end-user impact. It can also help convert natural language into NRQL queries to retrieve specific data points.
Conseil
Autopilot allows up to 100 requests per hour per organization.
Current limitations
Keep the following limitations in mind when using Autopilot:
- Autopilot recommends actions but does not take them. A person reviews and acts. Please review suggestions carefully for your use case.
- Autopilot does not create alerts.
- Autopilot does not instrument applications or generate instrumentation setup.
- Autopilot does not write back to external systems such as Confluence or ServiceNow. However, the Slack integration will allow Autopilot recommendations to be published to your Slack workspace.
- Investigation quality depends on your telemetry coverage and operational metadata. With limited data, recommendations may include less supporting evidence or identify additional information to collect.
- Investigation suggestions made by Autopilot depend on a multitude of factors, such as your telemetry coverage and operational metadata. With limited data, recommendations may include less supporting evidence or identify additional information to collect.