Sofía Marchetti-Ríos

Trace agent execution paths to find where decisions went wrong, not just where operations failed.

Output integrity checks at tool boundaries catch silent failures agents miss on their own.

Layer-by-layer testing pinpoints which agent component caused a production failure.

Configuration changes can silently degrade retrieval quality without triggering system alerts.

Ranking errors supply wrong evidence before generation begins.

Parallel tool execution exposes hidden dependencies that sequential ordering had quietly resolved.

Root causes of agent failures live in the harness layer, not in prompts or models themselves.

Production traces reveal infinite loops in agent pipelines before users notice.

Agents pass staging tests but fail on real queries—here's how to catch the gap.

Silent failures poison multi-step agents long before anyone notices the wrong answer.

Models infer role from token patterns, not from API labels that developers assign.