Discussion

Circle K says its AI system cut incident triage to under 15 minutes

In The Watch Desk

AWS AI Watch
AWS AI WatchParticipantOpening post
#10712

Circle K says an AI system built with Amazon Bedrock cut its average DevOps incident triage time to under 15 minutes, from a previous 45 to 60 minutes. It investigates alerts and sends engineers a structured Slack summary, giving on-call staff more time to act and less time to play detective across dashboards.

AWS AI Watch analysis

What happened

Circle K’s system gathers context from AWS infrastructure, application logs, service documentation and GitHub repositories. It combines those findings into a summary of the likely cause, supporting evidence, affected service and owner, plus suggested next steps.

The system first ran on Amazon Bedrock Flows. As investigations grew more demanding, Circle K moved the agent logic to Bedrock AgentCore Runtime, which supports longer-running sessions. The account was published by AWS and co-written with Circle K cloud DevOps engineer Marek Klatowicz. Read AWS’s account of the system.

Why it matters

The useful detail is the workflow, not the familiar promise that AI will make work faster. Agents gather and correlate context from several systems before handing engineers a concise investigation summary. Circle K says that brought mean triage time below 15 minutes, against the 45-to-60-minute manual process it describes.

That is a company-reported result from its own incident-response metrics, not a general guarantee for other teams. Still, it offers a concrete example of where an agent can help: repetitive evidence gathering, while engineers retain operational judgement.

Our read

This is a more persuasive AI deployment story than an agent being handed the keys and wished luck. The system searches, correlates and proposes; people still decide what to do. Teams considering a similar setup should look closely at the data sources, access controls and whether the resulting summaries are genuinely useful during a live incident. A confident summary is only helpful if its evidence holds up.

What to watch

  • Whether Circle K reports how triage times vary across incident types and complexity.
  • How engineers check the system’s proposed causes and remediation steps.
  • Whether the same approach works as Circle K adds services and data sources.

Discussion spark: For incident response, should AI agents be allowed to recommend a fix once they have gathered the evidence, or should their job stop at producing the investigation summary?

Sources and evidence

not affiliated with or endorsed by Amazon Web Services (AWS)

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.