Six months in, and I'm not sold. The promise is there, but the execution feels like a beta. Here's what I've seen.
The Good:
* The data collection is solid. Once configured, it's reliable.
* The UI is clean. Easy to drill down on specific traces.
* Cost is predictable, which is rare.
The Bad:
* The vendor benchmarks are suspicious. Claims of "zero overhead" don't match our internal tests. We saw a 3-8% latency increase in high-throughput services until we tuned sampling aggressively.
* The agent documentation is incomplete. Had to dig through GitHub issues to find critical config options.
* Alerting is basic. Threshold-based only, no real anomaly detection.
The Ugly:
* The "AI-powered" insights are mostly noise. Generic suggestions we already knew from looking at the charts.
* Support response time degrades after you're onboarded. Good luck getting a detailed answer on trace aggregation methodology.
Bottom line: It's usable for basic telemetry collection if you manage your own sampling. Don't trust their marketing on performance or insights.
Show me the methodology.
Spot on about the vendor benchmarks. We had the same experience with the latency claims, especially on our high-volume email routing service. That "zero overhead" line is marketing fluff for sure.
I'd push back a tiny bit on the AI insights, though. We found them useful for spotting odd patterns in lead source attribution we'd missed. But yeah, 80% of it was just noise telling us what we already knew from the dashboards.
The support drop-off after onboarding is real. It's like they forget you exist once you're past the initial setup. Makes me wonder if they're stretched too thin.
That support drop-off is a huge red flag for a platform like this. It screams they're focusing on new customer acquisition over retention, which never ends well for the core product stability.
The latency stuff is classic. We always bake a performance regression stage into our CI for any new observability agent. You catch that 3-8% hit before it hits prod. Makes the vendor claims a non-issue since you have your own data.
Interesting about the AI insights finding a needle in the haystack for lead attribution. Was that a pattern their system flagged, or did you just notice it because the dashboard made you look?
git push and pray