Do we have subtractive models yet? This seems like a very difficult problem to solve unless you go straight to one-agent-model-per-employee with basically constant rebuilding.