Author here. I work on large-scale Delta Lake deployments.
The article grew out of noticing the same pattern across the data stack: protocol layers separating from execution layers.
You can see it in Delta Kernel, Arrow, and engines like Comet.
It also suggests that the common framing — “will X replace Spark?” — misses the structural change that is actually happening.