SE Radio 719: Birol Yildiz on Building an Agentic AI SRE

Building agentic AI for site reliability engineering requires moving beyond static workflows toward reasoning loops that dynamically select tools to solve production incidents. Effective AI SRE systems prioritize "agentic search"—leveraging standard terminal commands like `grep` and `jq`—over heavy, high-maintenance vector databases to navigate large, complex data environments. By focusing on the "what" rather than the "how," developers allow reasoning models to determine the optimal path for root cause analysis, aiming to reduce resolution times from nearly an hour to under four minutes. Reliability in these systems is maintained through semantic evaluation pipelines and human-in-the-loop guardrails, ensuring that autonomous actions, such as patching or rollbacks, remain controlled. As AI-generated code becomes more prevalent, these agentic architectures provide a scalable, adaptable approach to managing increasingly novel and complex infrastructure failures.