OpenTelemetry Is Not Going Smoothly
The 'OTel isn't going well' thread surfaced frustrations that are widespread but rarely said plainly. The core complaint: OpenTelemetry is complex to instrument, documentation is inconsistent, and in practice it takes significant effort to get meaningful coverage. The alternative, paying for a vendor solution, trades complexity for lock-in and cost.
Some commenters pushed back with the obvious counterpoint: at least with OTel, you own the pipeline and can write your own exporters. Vendor solutions give you polish at the cost of control, and once you're deep in Datadog or Honeycomb's data model, you're not leaving easily.
This is a real founder trap. Observability looks like a solved problem from the outside and turns into a months-long instrumentation project from the inside. The OTel standard is winning the standards war but losing the usability war.
So what?
If you're building a product that needs observability and you're choosing between OTel and a vendor, be honest about your team's bandwidth to instrument properly. A half-instrumented OTel setup is worse than a fully-instrumented vendor setup. Pick the path you will actually finish.