What should people demonstrate at an internal AI showcase?
A showcase teaches the most when each presenter names the problem and the intended user, then walks a realistic input through the workflow. Include a hard case too, such as a wrong AI response. That's where other teams learn where the prototype belongs.
The culture panel described employees presenting working solutions to colleagues. That helps people learn from each other when it reveals more than an attractive final screen, so ask each presenter to say what result the prototype is meant to produce.
What happens when a source disagrees? A presenter who explains that limitation gives other teams something useful.
A working demo and permission to use it are separate things
A prototype may still need an owner, approved data access, support arrangements or additional review. Say so out loud. Give viewers a simple way to respond, like a similar problem on their team, an unanswered question or an offer to test with suitable material. Applause tells you very little about business value.
End with a named next decision: test further, adapt, pause or consider for an approved rollout. Record the reason. I'd count more testing as a perfectly good outcome, since sometimes it's the most responsible next step.