Visual thinking / Enterprise AI

The diagrams I use when AI conversations get too clean.

These are not universal laws. They are working models: compact ways to expose the constraint, ownership question or hidden cost that a capability conversation tends to leave out.

01Deployment stack 02Verification tax 03The intent gap

Model 01

The Enterprise AI Deployment Stack

Capability starts the conversation. The system around it determines the outcome.

AI value is produced through a chain: what the model can generate, whether people should trust it, who is accountable, and what actually changes. Most deployment failures happen above the model layer.

Questions it forces

What can the model reliably produce?
What evidence makes the output usable?
Who owns the decision and its consequences?

Model 02

The Verification Tax

The cheaper generation becomes, the more value—and cost—moves into verification.

A plausible answer is not a deployable answer. Enterprises pay for the evidence, review, controls and maintenance required to turn probabilistic output into something a real workflow can depend on.

Questions it forces

What must be checked before anyone can act?
Does the review cost erase the productivity gain?
Who keeps the ground truth current?

Model 03

Authentication Is Not Intent

Proving that an action was authorised does not prove that it was understood.

Identity establishes who or what may act. It does not automatically establish context, comprehension or intent. As agents execute more work, that gap becomes an operating and accountability problem.

Questions it forces

Who—or what—is taking the action?
What did the approving person believe would happen?
Where does accountability sit when intent and execution diverge?

A model should be useful, not merely neat

The point is not to simplify the organisation. It is to make the hard part visible.

See the thinking develop →