Why it matters
Untraced agent failures burn tokens silently. Repeated retries, runaway loops, and bad tool calls show up in traces before they show up on the bill, making observability core to spend control.
The tokenmaxxing angle
Converting traces into custom scorers is how teams find which prompts, tools, or retry patterns waste tokens. This workshop's core skill is root-causing agent spend, not just output quality.
From the organizers
Hosted by Braintrust at Four One Nine in SoMa, an intermediate-level workshop where attendees build a custom scorer using Braintrust's Topics feature, with a Granola representative as guest input.