Circle K describes AI incident investigation while engineers retain repairs
Circle K has described how its Amazon Bedrock system investigates production alerts and sends engineers a likely fault location and proposed remedy through Slack. A November case study covers the move from several agents to one persistent investigation session. The system gathers infrastructure, log and code context; engineers carry out fixes, with approval-based actions planned for later.
Artificial Intelligence··Midday
Alert investigations feed an engineer’s Slack summary
Circle K, the convenience-store and fuel retailer, described its AI-assisted incident investigation in a November 2 case study co-written with its cloud engineer Marek Klatowicz. Amazon Bedrock, Amazon's managed model service, brings together infrastructure status, application logs and code information. Engineers receive a Slack message naming a probable root cause, supporting findings, the service owner and a proposed remedy. They continue to implement repairs themselves.[1], [2]
One session preserves the investigation’s intermediate findings
Initially, separate agents checked infrastructure and logs before another searched the relevant code repositories. A final model call arranged their findings for delivery. As investigation context grew, the team moved the work into one agent using AgentCore Runtime, an Amazon service for running agents. That version entered production in April. Persistent session state keeps earlier findings available during later steps, and a gateway manages access to data-source tools.[1]
Approval-based repairs are a planned next step
The team now filters logs before supplying relevant lines to the model and uses those findings to narrow code searches. Its planned next steps include engineer-approved actions through Slack, such as restarting tasks or reversing a deployment. Connections to Jira and Confluence are also proposed. Today's system investigates and recommends; those action and context additions remain future work.[1], [2]
