Notes d’épisode
In this conversation, Kyle Forster, Founder of RunWhen, discusses what AI agents actually need to work safely in production. He explains why logs and metrics make up only around 30% of the data reaching an LLM during an incident, and why the other 70%, the CLI calls, API queries and dead ends agents accumulate between incidents, behaves like the knowledge a senior SRE builds over time. Kyle shares why typical stacks wreck accuracy, what "safe for production" means when agents have write access, and how the SRE role is changing with advancements in AI.

