Multi-agent architectures on the Databricks platform grew 327% in under four months. The dominant pattern: a supervisor agent breaking goals into sub-tasks for specialized workers. Organizations already knew how to decompose knowledge work. They were just waiting for the primitive.
That speed is remarkable. It also maps onto exactly the phase of industrial assembly-line adoption when the harder consequences started compounding. Taylor's decomposition boosted productivity immediately. Deskilling, quality erosion, management bloat arrived on the same timeline, just less visibly.
The honest read is that both things are true simultaneously. The ease of decomposition reveals something real about knowledge work's modularity. And the history of decomposition suggests the system-level costs are accumulating somewhere we haven't learned to look yet.