Who sets the rules in agent-to-agent negotiation? You do.
The biggest concern CFOs have about agent-to-agent negotiation is control.
Before any negotiation runs, the CFO sets the system boundaries — minimum return thresholds, days the agent never touches (payroll Friday, quarter close), whatever the rules of your business are. The agents operate inside those.
Control is not binary. Not every supplier gets treated the same way. Long-tail suppliers get handled autonomously. Strategic partners get a human in the loop: the agent surfaces its proposed move, you approve or adjust, it sends. Over time, as trust builds, you widen the boundaries where it makes sense — "allow once" gently becomes "always allow."
The next question is: what happens on the other side? If my agent has guardrails but my supplier’s agent is optimized to be accommodating, do they just agree their way into a bad deal? Salesforce AI Research calls this failure mode "echoing" — two agents optimizing for agreement instead of outcome.
The answer is the same: the agent has a floor. It does not negotiate below your minimum return, no matter how reasonable the other proposal sounds. It is not trying to reach agreement — it is trying to reach the best outcome within the constraints you gave it. With the right guardrails, the agent is always protecting your interests.