With all due respect, that’s a blatant misunderstanding of gradle (though gradle’s documentation not being clear enough, giving to the spread of this misunderstanding is a fair criticism).
Gradle is not too powerful, it builds a build graph based on the .gradle file, and the building itself happens imperatively — but the actual execution is static, based on that fixed build graph. This configuration is cached and subsequent executions will all refer to this static graph to know what requires rebuilding.
This is the minimum required complexity/generality for a build system that is not “hardcoded” for a given task (e.g. to a degree maven, but cargo, go’s build tool etc all can only compile their respective languages on their own. “Plugins” can let them do more, but at that point plugins have to re-develop all the functionality of a build system - caching and parallelization), and as I mentioned there really are only a handful of tools capable of that.
As for valid criticisms of gradle - it’s API historically experiencing constant breakages, used to be dependent on JDK version (and lagged behind for a long time due to Groovy), which is more or less fixed now with toolchains, and the model not being clear enough for people not wanting to dive deep. E.g. I can count on a single hand how many people knows that doFirst/doLast should be used for the actual task itself, anything else will happen at the config build time which is not what you want.