AI-assisted incident investigation.
AI SRE agents correlate logs, metrics, traces, and runbooks into an evidence-backed investigation your on-call engineers can verify and act on.
| Good for | SRE, platform, and on-call teams |
|---|---|
| Access | Read-only to start |
| Engagement | Scoped project or $200 / hour |
| First step | A short brief by email. Free. |
What the work involves
AI SRE agents support site reliability engineering (SRE) and operations teams with service-health questions and incident investigation. They query approved tools, gather the relevant context, and explain their findings with links to the evidence.
- Connect approved observability sources, service documentation, and recent deployment context.
- Investigate a specific alert or service issue across logs, metrics, and traces.
- Surface relevant runbooks, likely causes, and missing information for an engineer to review.
- Prepare an investigation summary with supporting evidence and suggested next checks.
When it helps
Engineers spend time switching between monitoring tools and documentation to understand incidents, and you have accessible telemetry and past investigations to evaluate the agent against.
At Macy’s I built AI agent skills for observability and SRE workflows, supporting troubleshooting and daily engineering operations.
A typical project
An API starts returning errors after a release. The agent checks the error window, inspects the relevant traces, and compares recent changes. The engineer gets a short, evidence-backed summary, a clear statement of what’s still uncertain, and the suggested next checks.
What you get
| Deliverable | What it includes |
|---|---|
| A focused investigation workflow | One defined service or incident type, connected to the agreed data sources with scoped access. |
| Evidence your team can verify | Findings linked to source data, with uncertainty stated plainly and a handoff to the engineer. |
| Evaluation and operating guidance | Known incident scenarios, checks for unsupported conclusions and missed signals, and documentation for monitoring and maintaining the agent. |
Questions
Will the agent make changes in production?
Not in the initial scope, which is read-only investigation. Engineers review the findings and decide what to do. Remediation can be scoped separately, with explicit approvals, limited permissions, and a rollback plan.
Can it work with our current observability tools?
Connections are built for the tools you already use, based on their APIs and permissions. A first pilot can start with one or two sources, adding more once the investigation workflow has been evaluated.