Free Lesson
Building Your Agent Indexing System
60 min
Sep 25, 2026 1:00 PM
Virtual (Zoom)
In this video
What you'll learn
How observability standardized, twice
Metrics consolidated on Prometheus. Tracing consolidated on OpenTelemetry. Both times the fix was an agreed format.
The terms of art: trace, span, metric
The vocabulary you need to direct an agent to build this, and the naming conventions that make harnesses comparable
Where the span model stops
Spans assume short, thin, and disposable. An agent session is hours of prose you want to keep and search for years
Why you want warehouse storage
Keep the raw capture beside a versioned Parquet view, so a schema change becomes a rebuild instead of lost history
Hybrid search over a real trace archive
Live CLI and UI over six months of my own agent sessions, and what the index actually costs per month
Why this topic matters
A coding agent reads more of your repository in an afternoon than a new hire reads in a month. It writes down every file it opened and every approach it abandoned, and then the session closes and all of it goes in the bin. What survives is usually run counts and token spend, which is enough to bill against and too thin to engineer against. This talk builds the archive instead.
