#agent-observability

What OpenAI, Anthropic, Google DeepMind and Meta each publish on whether an agent's reasoning trace can be read and trusted: every measured figure, every blank.

eve 0.44.0 gates trace creation by conversation audience: only public is retained by default, while private and unknown conversations fail closed.

Codex's MultiAgentV2 now encrypts what a parent agent tells its subagents, so developers lose the local audit trail. Why the July 15 disclosure matters.

Anonymised LangFuse rollout for agent observability — situation, approach, OpenTelemetry adoption, cost attribution, drift detection, outcomes.

Roll out agent observability in 90 days — vendor pick, trace coverage, eval signals, drift detection, replay infra. Phased timeline with template kit.

Eight observability anti-patterns that make agent traces useless — PII in spans, cardinality explosion, trace truncation, missing replay, more.

Per-trace cost calculator at 1K, 100K, 10M monthly volumes across LangSmith, LangFuse, Helicone, and Phoenix — feature deltas and break-even math.

Audit agent observability across 60 points — trace coverage, span depth, eval signals, drift detection, cost tracking, and incident response readiness.

Six agent-observability platforms compared on tracing, eval, cost, and production-monitoring. Workflow examples + decision matrix for agency stacks.

Defining ASR for production agents — completion vs partial vs hallucinated, cost-adjusted scoring, and benchmark suites. Formal methodology + reference dataset.

Agent observability guide — LangSmith, Braintrust, Langfuse compared, eval patterns, trace sampling, and cost attribution for multi-tenant agents.