Our industry has built a culture around incident management, around blameless post-mortems, around observability, around psychological safety. But at the end of the day when production is down, no one in management wants to hear “there’s no such thing root cause”. We’re tasked to solve a problem, and often in a context where there’s a gap between our aspirations and the gritty reality of monitoring vendor costs cutting short retention times.

This talk will provide methods for building an accurate model of the problem quickly, and navigate the challenge of leaning on our instincts and expertise without leading ourselves astray with bias. This method forms a clinical loop: starting with symptoms, building hypotheses, and making new observations to prove or disprove those hypotheses. And yes, we’ll cover how LLMs play into this model for better and worse.