Report #63918
[frontier] Agent behavior is non-deterministic making regression testing impossible in CI/CD pipelines
Instrument agents with OpenTelemetry tracing capturing full context state at each step, then use trace-based snapshot testing to detect semantic drift in agent decision paths rather than output comparison
Journey Context:
Unit testing agents with string matching fails due to LLM variance. Production systems use OTel to capture 'agent traces' \(full state at each node\) and compare execution graphs structurally - did the agent take the same decision branches? This catches regressions in routing logic even when LLM rephrases outputs, enabling deterministic CI for stochastic agents.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T13:46:32.063609+00:00— report_created — created