The world is moving fast, and the ground is shifting beneath our feet. Yet over the last decade. observability has meant the same three tools for three data types: metrics, logs, and traces. It's a framing sold by vendors, and it forces you to spend time stitching context across three systems, three query languages, and three retention policies every time production breaks. This talk makes the case that the industry is converging on something better: the wide event. One structured, high-cardinality record per unit of work, from which metrics, logs, and traces are derived at query time. We'll cover why column-oriented storage makes this affordable, why it's the only viable model for LLM observability, and what an eval harness revealed about agentic root-cause analysis on wide events versus the three pillars.
The world is moving fast, and the ground is shifting beneath our feet. Yet over the last decade. observability has meant the same three tools for three data types: metrics, logs, and traces. It's a framing sold by vendors, and it forces you to spend time stitching context across three systems, three query languages, and three retention policies every time production breaks. This talk makes the case that the industry is converging on something better: the wide event. One structured, high-cardinality record per unit of work, from which metrics, logs, and traces are derived at query time. We'll cover why column-oriented storage makes this affordable, why it's the only viable model for LLM observability, and what an eval harness revealed about agentic root-cause analysis on wide events versus the three pillars.
Observability Nerd
Jordan Simonovski has historically spent way too much time scaling observability systems to their limits. He's currently building the future of observability on ClickStack at ClickHouse, and is currently on a mission to kill the three pillars of observability and pushing people toward systems that make more sense in today's AI-driven landscape. He was a technical reviewer for Observability Engineering, 2nd Edition, and has OpenTelemetry running in his espresso machine. You may also know him as the person behind Simpsons Against DevOps.
