I was arguing against the notion that it necessarily leads to a combinatorial explosion. But there is a definitive cost to using feature flags, which I believe can be weighed against other options. In the types of environments where I work, we are able to avoid spending this cost most of the time, without major drawbacks. I'd say, the rare feature flags we have are very short-lived and mostly in the frontend (spa), very rarely in the backend, probably due to a decent team discipline around backward compatibility.