Some will argue that the answer is better feedback loops, not fewer agents. And they are right, up to a point. Feedback loops are essential. But feedback loops, to be effective, are themselves deterministic by design: If this input, then that correction. Without deterministic guardrails, a feedback loop is just another probabilistic suggestion. The agentic future we are being sold rarely includes those guardrails, because they sound boring. But boring is what keeps the lights on.
This is the ethical innovation challenge I want to leave you with. We are rushing to deploy agents that are brilliant at probabilistic variation, at creative, unpredictable problem-solving, into environments that demand deterministic reliability. Payroll must be paid. Regulatory filings must be accurate. Safety checks must be completed identically every time.
The vendors will tell you that their agents can handle both. They cannot. Not yet. Agentic AI is wonderful for discovery, for surfacing unexpected connections, for the fuzzy front end of innovation. But for the deterministic backbone of your business? The rules-based processes that keep the lights on?
Stick to the scone recipe.
As KM professionals, your job is to be the gatekeepers of this distinction. Don’t let the probabilistic hype erase your deterministic reality. Measure expertise by outcomes, not uploads. Demand that your AI vendors tell you, explicitly, where their system is deterministic and where it is probabilistic. And if they can’t answer? Smile, nod, and walk away. Your audit trail will thank you.