What happened
Datadog released LLM Observability 2.0 on August 22 with end-to-end agent trace visualization, per-step cost attribution and anomaly detection for tool-call loops.
Integrations cover OpenAI, Anthropic, LangChain and LlamaIndex. Datadog said beta customers reduced runaway agent spend incidents by 50%.
Why it matters
Agents fail in loops and silent cost spikes that traditional APM misses. Agent-specific traces and cost attribution give platform teams data to tune prompts and routing.
Anomaly detection on tool loops catches a common production failure mode early.
MintedBrain take
Instrument agents before scaling users, not after the first billing surprise. Trace data also helps debug quality issues faster than replaying logs by hand.
Discussion
Sign in to comment. Your account must be at least 1 day old.