Too Many AI Demos Fail to Show the User - And the Cost

“Hey… you wanna see something really scary?”

That line from 𝘛𝘩𝘦 𝘛𝘸𝘪𝘭𝘪𝘨𝘩𝘵 𝘡𝘰𝘯𝘦 often comes to mind when I see an AI demo.

AI demos are everywhere right now.

And don’t get me wrong, many of them are impressive.

• They show agents completing tasks.

• They show models summarising information.

• They show content being generated in seconds.

• They show automation replacing manual effort.

But too many AI demonstrations still fail to answer two critical questions:

Who is actually going to use this?

And:

Who is responsible for understanding what it costs every time it runs?

That’s really scary.

The value of AI is not established simply by proving that technology can perform a task.

It is established by showing that a person can use it in a way that makes their work meaningfully better, and that the organization can operate that capability responsibly at scale.

An AI demo should therefore show more than “Here is what the AI can do.”

It should show:

• What triggers the process?

• What does the AI return?

• What must the person review or validate?

• What decision can they make next?

• What work has actually disappeared?

And most importantly:

Is this genuinely easier than just doing the task themselves?

AI can automate part of a task while introducing new work around it: prompting, checking, correcting, approving, reformatting and moving information between systems.

If the demo does not show the human workflow around the AI, it is difficult to know whether work has been removed or simply moved somewhere else.

There is another reason personas especially matter in AI demonstrations:

Each AI usage has an operational cost.

Every model call, agent execution, reasoning step or automated workflow consumes something: tokens, credits, included usage or metered charges.

And with increasingly agentic systems, one user request can trigger multiple downstream executions before an outcome is produced.

So personas should not only be shown using the AI.

They should also be shown operating it responsibly.

• Who can see usage accumulating?

• Who understands the cost of an execution?

• Who is alerted when consumption changes unexpectedly?

• Who can constrain or stop runaway agentic activity?

• Who is accountable for keeping usage within agreed limits?

That is how an organization constructively begins a discussion about the potential value of an AI product.

Not because the AI did something technically impressive.

Because you can see how a person uses it, what changes in their work, what it costs to operate, and how that person remains accountable for its use.

The AI is not the story.

The person using it - and responsibly operating it - is the story.