Your agent already told you what to evaluate
How Triana Preview turns authorized production traces into a private, evidence-linked Behavior Map for evaluation planning.
Read article →Triana Engineering
Evidence from building safer AI agent releases.
Experiments, failure analysis, and practical eval engineering for conversational and tool-using agents. Written from the traces, not from a content calendar.
RSS feed ↗Latest
The score improved. The failures migrated: what aggregate AI agent evals hide
An engineering study of how an eight-point improvement in an AI agent evaluation score concealed a substantial migration in failure behavior.
Read article →