Your AI DBA Will Hallucinate During an Outage

This talk brings together Fernando Ipar, DBRE at Life360, and Andrew Morgan, CTO of Sixta, to share hard-won lessons from a year of building and running AI-assisted database incident investigation in production. We’ll cover the design tradeoffs that determine whether an agent actually helps during an incident: how you feed it context, what you let it reason about versus what you keep deterministic, and how you build in the institutional knowledge that separates a useful investigation from a confident hallucination. We’ll talk about where the agent genuinely earns its keep, and we’ll be honest about what still breaks. Expect candor, architecture decisions you can steal, and a realistic picture of where this technology actually is today.