There is no intelligence without hallucination. So what do we do about it?

Agentic AI drift and hallucination will always be a thing to audit for, for the same reason human drift and hallucination will always be a thing to audit for.

We like to point the finger at AI as especially foolish or malicious when it comes to hallucinations and made-up facts. And yet I challenge you: for the next week, pay special attention to your conversations with friends, family, and colleagues.

Make an honest self-check when you hear claims of fact coming out of your mouth or theirs. Truly, what evidence do you have to support that statement? How sure are you that you (or they) are correct? Have you fact-checked it? Notice the overall tenor of these conversations as claims are thrown back and forth.

The same confidence with which we throw around unfounded claims: you'll find the AI agent doing exactly the same. Humans do it perhaps due to emotional drives; agents do it according to their probability landscapes. But we all do it just the same.

Agentic intelligence and human intelligence have something important in common: they are both non-deterministic.

What is non-determinism? It describes any system where the same input can produce different results.

Every non-deterministic system needs to be set up with sufficient "deliberative friction" to protect it from delusion.

This is why we have 12 jurors, 9 justices, panels, councils, and deliberative bodies of all kinds. Though often frustrating to those who just want a decision made, the friction is the value in these processes.

The worst danger for a leader is to be surrounded by "yes men" who confirm every hallucination and delusion without integrity. The fable of "The Emperor's New Clothes" comes to mind. What is agentic drift but the courtiers surrounding the emperor?

For human teams, process tools like retrospectives, ensembles/mobbing, and chartering are crucial for the kind of friction that creates higher-quality decisions. There's no way around this need to peer-audit.

In ClickChain's governance, we have embedded these same deliberative frictions into our agentic processes. It is one essential element of why we can guarantee safety and quality. Not because the drift disappears, but because the friction catches it.

If you remember nothing else I've said about AI here, just remember: intelligence means the capacity to drift. So what are you going to do about it?