telemetry.dev
I keep coming back to the same question when I see agentic AI: how do you actually know what it's doing? Telemetry.dev is answering that with observability built specifically for LLM agents, and that's a real gap right now. Traditional monitoring tools were designed for predictable code paths, not for models that can take unexpected detours mid-task, hallucinate a step, or just silently burn tokens on a loop.
The timing makes sense. Agents are leaving the demo phase and hitting production workloads where a single bad step can cascade into a mess. Teams need to trace decisions, see token usage, and spot exactly where the model went off the rails. If telemetry.dev can make that as easy as checking a dashboard, it has a shot at becoming a standard layer in the stack rather than a nice-to-have.
I'd want to know how deep the instrumentation goes and whether it handles the messy reality of multi-step agent workflows without drowning teams in logs. But the direction feels right. Observability was a huge deal for microservices, and agents might need the same treatment, just with a lot more context about intent and reasoning baked in.