Here’s an interesting new feature from Elastic that gives users the ability to see OTel GenAI traces from agents. Coincidentally, I wrote about this in August 2026 (see Why AI Agents Need Standard Observability).

Found at “LLM tracing in Elastic APM: prompts and token counts” on 2026-09-04T13:53:31+02:00.

Elastic APM now does LLM tracing in the trace view. The GenAI tab in the span flyout has the whole conversation, so you can read the system prompt, the user messages and the model response, and copy any of them. Every GenAI span row in the waterfall shows input and output token counts, so in an agentic trace with twenty LLM calls you can find the span using the most tokens without opening any of them. Your LLM calls are now in the same waterfall as your database queries and HTTP spans.