I haven't opened a dashboard in six months

I don't think I have opened an observability dashboard in six months. Not because I don't need one: because Claude takes care of it, and does it better than I do.

Over the last two years I have worked with quite a few clients on observability infrastructure built on open tools. The starting point has always been the same: OpenTelemetry for collecting signals. I'm convinced OTel reached, some time ago, a level of maturity and completeness that makes it the right choice for logs, traces and metrics, with the non-trivial bonus of being able to switch backends without re-instrumenting anything.

As a backend, my current default is ClickStack. I like it because it keeps logs, metrics and traces in a single database (ClickHouse), without having to glue together three different products and three different ways of writing queries.

That said, the interface itself (ClickStack's UI is called HyperDX) matters less and less. I no longer find myself opening a dashboard, writing a query, jumping from one chart to the next to figure out what is going on.

What matters is that the data is well organised in a suitable database. From there, the agent takes over. The only requirement is that its harness has adequate access to the observability data: ClickStack, for instance, exposes an MCP server.

When something breaks, Claude works out how to use the tools, runs the queries it needs and gets to the cause. It is generally faster and more effective than I would be doing it by hand. By a lot.

I don't know yet what this means for the future of observability consoles. I do know that, for me, the visual part has become a detail.