Full disclosure: I am a vendor in the space. But, the most "awesome" part that we've seen is when we bolt AI capabilities on top of the data. Instead of the interface being a relatively static React app, the interface becomes conversational and directly in your IDE.

There are certainly solid areas of Backstage, but they tend to be found in the core path: think docs, scaffolding, etc. and less so in the plugin ecosystem. Plugins have been a struggle for the community due to varied support levels, quality concerns, etc. Oftentimes plugins are brought to the community to be able to mark the "we support Backstage" check box, but are not always well-maintained after that.

For us, we were always after the data that exists under the IDP...essentialy a graph of the entities within the org. So, for us, the watershed moment was using that data to connect the entity data with an LLM (and other tools like MCP) to provide a natural language interface. From that interface we can interrogate and take specific actions against the live systems expressed by our entity graph.

e.g. like you're calling http apis or something?

Backstage and similar tools will give directed views of ie. last Actions run status. But here we are able to use the LLM to create more complex flows. For instance: "tell me which repos have the highest rate of Actions build failures"

So here instead of looking at a (mostly) static React site which probably require the end user to dig for the data, I am asking the tools to correlate the data for me.