AI SRE: triage, noise and investigation
Noci AI does the work of an AI SRE. It triages every alert as it arrives, alongside paging. Each alert gets a noise score with the reasons shown. Agents then investigate with your telemetry and recent code changes, and show a hypothesis with the evidence behind it. You choose, step by step, whether an agent asks before it acts.
What does an AI SRE do during an incident?
An SRE spends the first minutes of an incident gathering: open the alert, find the dashboard, read the logs, check what changed. An AI SRE does that gathering as soon as the alert arrives, and puts the first reading in front of the person who is paged. Noci AI does three things in that order.
- Triage. Every alert is read as it arrives and scored.
- Noise. The score comes with its reasons, and noisy alerts can be kept off an escalation policy.
- Investigation. Agents look at your telemetry and recent code changes, and write down what they found.
How does triage work with paging?
Paging goes out at once and AI triage runs alongside it from the first alert. If you would rather interrupt people only for the alerts triage lets through, you can have paging wait for triage. The choice is yours, not the tool’s. The longer version is in alert triage for SRE teams.
How is noise handled?
Each alert carries a noise score and the reasons for it, shown beside the alert. You can keep alerts above a flakiness score off an escalation policy. The alert still shows its score and reasons, so a limit that is wrong can be found and fixed. Noise is half of the loop that leaves a company with unwatched services: when people drown in alerts they mute them. The other half is blind spots.
What does an investigation produce?
A hypothesis, and the evidence behind it. Agents read your telemetry and your recent code changes, line them up against the time the alert fired, and show what they checked. They do not declare the cause: a person reads the evidence and decides. See AI root cause analysis for how that works in practice, and what an AI SRE is for the wider idea.
What the agents read is listed on the Context Lake page. What they may do next is yours to set on the agents and actions page.
Questions
Does triage delay the page?
Only if you choose it to. By default the page goes out at once and AI triage runs alongside it from the first alert. If you choose, paging waits for triage, so the person paged starts with a score and the reasons.
Can I see why an alert was scored as noise?
Yes. Each alert carries its noise score and the reasons beside it, so a person can read the decision and challenge it.
Does the AI decide the cause of an incident?
No. Agents give a hypothesis and the evidence for it. A person reads the evidence and decides.
What does an investigation read?
Your telemetry and your recent code changes, from the tools you connect. The Context Lake page lists what Noci AI reads.