Skip to content

CONCEPT Cited by 1 source

Reasoning trace

Definition

A reasoning trace is a persistent artifact capturing the decision-making process of an AI agent during a single execution pass — which tools were called, in what order, what evidence was gathered, what the agent considered, and what it concluded. Distinct from the final output (which is what the user sees), the trace captures how the agent arrived there.

In Instacart's Blueberry, every diagnostic pass writes both: 1. A session-context snapshot — for the next pass or a different worker to take over without losing thread state 2. A reasoning trace — for developers to later inspect how the agent decided what to do

"These artifacts are what make the rest of the learning loop possible." (Source: sources/2026-07-14-instacart-blueberry-on-call-reasoning-harness)

Why it matters

  • Debugging: when the agent produces a wrong explanation, the trace shows which evidence it weighed and where the reasoning went wrong
  • Improvement loop: traces surface patterns in agent behaviour that inform skill/tool refinement
  • Trust: operators can verify that explanations are backed by real evidence, not hallucinated

Seen in

Last updated · 585 distilled / 1,765 read