Sam Alba, co-founder of Mendral and early Docker/Dagger contributor, discusses building an AI DevOps engineer agent that monitors GitHub Actions CI/CD pipelines, detects issues like flaky tests, slow builds, security misconfigurations, and linting failures, then proposes or automatically implements fixes. The conversation covers Mendral's architecture — ingesting CI logs and events, maintaining a living memory of insights, and operating as a Slack-integrated agent that asks for permission before opening PRs. Key themes include the non-determinism argument for AI in ops (humans are also non-deterministic), tribal knowledge capture via agent memory, progressive trust-building toward automated PR merging, and the growing pressure on DevOps engineers to adopt AI tooling as coding agents dramatically increase PR volume. The host shares hands-on experience managing 100+ repos and how Mendral helps prioritize and triage CI failures at scale.