How to tell AI washing in a vendor demo

Direct answer: AI washing in a vendor demo is awe without accountability: no named job, no failure mode, no data boundary, no owner, and no result you can reproduce on your own redacted data.

What this means at work

Demos are optimized for applause. Workflows are optimized for Tuesday. If the pitch never says what breaks when the model is wrong, you are watching theater.

Score the demo on paper before you change a process. Admiration is allowed. Adoption is earned.

Do this next

1. Ask which recurring job this replaces or assists in one sentence.
2. Ask what happens on a confident wrong answer.
3. Ask where your data goes and what must never be pasted.
4. Ask who signs off in your org. If “the AI” is the owner, stop.
5. Demand a redo on your redacted sample, not the keynote clip.

Write the four answers down. If the vendor cannot help you fill them, you learned something valuable: the demo was not ready for your Tuesday.

FAQ

Is every polished demo washing?

No. Washing is polish that dodges job fit, risk, data, and ownership.

What if they refuse a redo on our data?

Treat the capability as unproven for your workflow.

See also

• Pulse: Score the demo before the workflow
• Challenge: Surviving the hype cycle
• Field Guide: Verify