Most of us already know that AI is inside the organisation.

Employees are using it. Executives are using it. Vendors are embedding it into existing products. Teams are building with it. And increasingly, AI is shaping the information people use to make decisions.

But some organisations still govern AI as though it were conventional software.

In this episode of Decoded: The Decision Layer, Joel Leslie explores why that is no longer enough.

We look at what changes when AI moves from generating information to recommending, influencing and acting; why instructions are not the same as controls; why the falling cost of building software can hide the true cost of owning it; and why organisations need to think carefully before deciding to build AI products themselves.

We also examine a deceptively simple CSIRO case showing how an AI agent can follow an instruction while completely missing the intent behind it, and what that means when the same behaviour moves into claims, payments, procurement, workforce decisions or customer service.

And with Australia’s new automated decision-making transparency obligations commencing on 10 December 2026, there is another question leaders should be asking now:

Could you identify where AI is already influencing consequential decisions inside your organisation?

Because AI assurance is not a one-off exercise.

AI assurance has an expiry date.

The goal is not to slow adoption. It is to build the governance, ownership and assurance needed to use more AI safely, productively and with confidence.

Episode resources and further reading:
DecisionLayer-Podcast.com