Circle K now diagnoses IT incidents in under 15 minutes, down from as long as an hour.

Circle K, a convenience store operator with more than 17,200 stores across 27 countries, has moved GenAI incident triage into everyday production use. The company described the system on 2 November 2026, in an account co-written with Amazon Web Services. Mean triage time has fallen from 45 to 60 minutes to under 15 minutes, a reduction of 67 to 75 percent, measured against the company’s own incident response records.

What GenAI Incident Triage Replaced

Before the agents existed, the first stretch of every incident was manual. An alert told the on call engineer that something had broken, but not what or why. Finding out meant checking cloud resources, searching application logs and traces, and reviewing recent code changes. Mean triage time ran 45 to 60 minutes. Complex incidents spanning several services took more than an hour.

Circle K built its GenAI incident triage on Amazon Bedrock, Amazon’s managed service for running foundation models. When a monitoring alert fires, software agents now carry out that investigation instead. They inspect the affected cloud resources, query the logs and traces, and check recent code changes. Then they post a written summary into the engineering team’s Slack channel. As a result, the on call engineer opens a diagnosis rather than a blank page.

The first version spread the work across three separate agents. In April 2026, however, the team folded them into a single agent with brokered access to each data source. That design now runs in production and handles the full range of incidents, from single service alerts to large multi service investigations.

The Result and What It Signals

Straightforward incidents now receive a summary in five to eight minutes. Complex multi service incidents, which previously took more than an hour to triage by hand, now finish within 15 minutes.

Notably, the gain does not come from better judgment. It comes from removing the evidence gathering that used to sit between an alert and a human decision. The agents do not resolve the incident. Instead, they hand the engineer a starting point. Other operators have reported similar compression in the same stage of the work, including GoTyme, which dropped root cause analysis to 10 minutes.

Why It Matters

  • GenAI pays off fastest where the bottleneck is gathering context, not making the decision. Incident triage, claims intake and case preparation all share that shape.
  • The signal to watch is how much of an expert’s time goes to assembling information before any judgment is possible. Where that share is high, agents have room to work.
  • Consolidating several narrow agents into one, as Circle K did, suggests the first architecture is rarely the final one. Budget for the rebuild.
  • Outputs that land where people already work, such as a team chat channel, avoid the adoption problem that separate dashboards create.