Rescale has described an AI test-triage agent that automatically classifies failures from its nightly end-to-end test suite and routes them for follow-up. The company says the system was developed from initial problem definition to staging deployment in five days and reached production within weeks.

The agent runs as a step in Rescale’s existing Argo Workflows CI/CD pipeline on Amazon Elastic Kubernetes Service. According to the company, it uses Claude Opus 4.8 through Amazon Bedrock to categorize failures as infrastructure issues, code bugs, flaky tests or environment problems. It also retrieves historical failure patterns, queries pod logs through the Model Context Protocol and produces structured remediation reports with file references, function names and recommended fixes.

Rescale says the agent was built with the Strands SDK and was designed to support concurrent failure analysis. Engineers review the reports before any action is taken. The company also created a PyTest-based evaluation framework with live and stubbed test modes to measure classification performance.

In its account of the project, Rescale says agreement with human triage rose from 44% at initial deployment to 68%, and that the system has contributed to a 20% decrease in the time production deployments are blocked while failures are analyzed. These figures are company-reported. Rescale says the agent is now running in production and that the approach is being extended across its engineering organization.