01
What debugging agents need from observability SDKs
A source-based comparison of Foam’s Node and Ruby packages with OpenTelemetry, Sentry, Datadog, New Relic, and Honeycomb.
Foam Research · 2026
Research for systems that have to work.
We study how agents investigate production software: what context they need, which tools help, and how to measure whether they found the actual cause.
Marginal tool utility in agentic debugging
More tools do not necessarily make an agent better. Removing noisy tools preserved accuracy while doubling the share of calls that made useful progress.
Tool efficiency
Default suite0.359
Signal only0.720
2× efficiency · accuracy preserved
Publications
05 records02
Clustering and noise reduction
How streaming template mining compares to Sentry fingerprinting across 11 production environments.
03
Why AI SREs Make No Cents.
AI SREs need complete telemetry, but legacy observability pricing makes every additional byte of context more expensive.
04
Structural gaps in error monitoring
Evidence and causes behind failures in grouping, prioritization, configuration, alerting, and AI-generated fixes.
05
Root cause accuracy from observability data
A benchmark for the question every debugging agent should answer: what caused the production failure?