I reached 1,000+ telemetry events across 38 deployments — what should I be thinking about now?

I've been building an open-source AI workflow automation platform for a while now, mostly as a solo project. I recently checked the telemetry dashboard and was honestly surprised to see 1,062 telemetry events across 38 unique platform deployments. The project has gone from a relatively small workflow runner to something with workflows, branching, tools, agent memory, document RAG, scheduling, webhooks, and a visual workflow builder. I'm now at the point where I'm starting to think less about “what feature should I build next?” and more about “how do I avoid breaking things as more people actually use this?” For people who have maintained open-source projects that started getting real usage: What did you wish you had put in place earlier? Things like: How do you handle backwards compatibility when changing the workflow format? When do you start worrying seriously about migrations? How do you safely roll out updates without breaking existing installations? What kind of telemetry/observability is actually useful? How do you deal with bug reports from environments you can't reproduce? Are there any mistakes you made when your project went from “a few people trying it” to actual deployments? I'm particularly interested in lessons from people maintaining self-hosted/open-source infrastructure, since users control their own installations and I can't simply push an update to everyone. Would really appreciate hearing what you learned the hard way.