In this episode of Google's podcast about Site Reliability Engineering (SRE) and production software, host Steve McGhee, along with Matt Siegler, interviews Todd Underwood, Head of Reliability at Anthropic, about the current and future applications of AI and ML in SRE. They discuss the hype around AI, its limited success in areas like anomaly detection, and potential in coding, config generation, and documentation interfaces. Todd emphasizes the importance of end-to-end model quality as an SLO and the need for responsible AI development, citing Anthropic's responsible scaling policy. The conversation touches on the balance between product velocity and reliability in AI releases, the potential harms of rapidly deployed AI, and the role of SREs in ensuring AI safety and usefulness. Todd shares his surprise at the real market for AI services and the fascinating potential of AI agents, while also cautioning against unfounded hope and advocating for thoughtful AI development.
