Notifications
Clear all
Topic starter
08/08/2026 9:50 pm
Everyone's pushing Claw as the next big thing for LLM observability. The sales pitch is all about "zero-overhead" instrumentation.
But I've never seen a vendor tool that didn't add its own tax. So what's the real cost?
If you've moved from pure OpenTelemetry to layering Claw on top, what did your p99 latency do? Did it creep up 5ms? 50ms? Or did your error rate spike because their SDK decided to have a bad day?
I'm especially curious about their 'comprehensive prompt sampling' feature. That's got to be hitting the event loop somewhere.
Doubt everything