Drift itself is not loud at all. What happens is that a model gets some information and then a threshold moves. This means a rule for routing changes. Things do not break The behavior changes over time.
With Fabric Foundation and ROBO the question is not if agents act. If their behavior stays the same even when things change all the time. Drift makes things worse. Each logical change makes a system that no one planned out. We can see the cost in things like ,variance policy half-life and reconciliation load. This means that outputs do not match up rules change in ways that're not clear and operators have to spend more time fixing the differences.
You can use ROBO tokens to pay for tools that let you see and replay things. Autonomy does not usually fall apart at once or it slowly fades away when people are forced back into the loop nothing goes wrong it just gets harder to trust ROBO and the system and that is more expensive than failing.