What does an employee need before they can experiment with AI at work?
Telling people to experiment doesn't tell them which tool to open, what information is allowed or when a test needs approval. Sejal Amin described making the easy path the safe path for teams running experiments, and that's a design job. The route has to be something a person can follow.
Spell out an approved starting route in terms someone can follow. Name the tool and where it runs. Say who has access and where practice material lives. Check the tool's permitted use with the responsible teams, since an available login doesn't mean every experiment is covered.
Say up front whether a test can use fictional cases or read internal documents, and whether it can send messages or change records. A practice task that produces a local draft is a different animal from one that acts on a live account.
People also need somewhere to take an uncertain case. Name the person or team who answers, and what information to bring them. If every question lands in an approval queue nobody owns, there's still no usable route.
Plan for a test that misbehaves, too. Employees should know how to stop it, keep the evidence and report the problem without exposing more private information.
Last, give the instructions to someone who didn't write them and watch where they hesitate. I'd want them to leave knowing how to begin and where their permission ends.
What to try this week
Pick one experiment people already ask about and write a half-page route for it: the tool, who can use it, what information is allowed and who answers questions. Post it where the team already looks.