How to Get a Team Confident With AI

Most teams that "are not using AI" are using it. Quietly, on personal accounts, without telling anyone — because nobody said whether it was allowed, and asking felt risky.

That is the real blocker. Not capability. Permission.

Start by saying what is allowed

Before any training, answer three questions in writing, in plain language:

• What are we encouraged to use AI for? With examples from your actual work.
• What must never go into a public AI tool? Client data, health information, anything under NDA — with concrete examples, not categories.
• When must a human check the output? Anything client-facing, anything with numbers, anything you would be embarrassed to get wrong.

Half a page. Circulated. That single act unblocks more adoption than any training session, because it converts an ambiguous risk into a clear boundary.

Train on their work, not toy examples

Generic AI training fails in a predictable way: everyone enjoys the session, nobody changes what they do on Monday. The gap is that the examples were not theirs.

Have each person bring one real task they do weekly and dislike. Work on those in the room. People leave with something that already works on their own job, which is the only thing that survives contact with a busy week.

Name the failure modes out loud

A team that has only seen AI work well will trust it in exactly the wrong moment.

Show them a hallucination. Show a confidently wrong answer on something they know well. Explain that a public tool may retain what they paste. This is not fear-mongering — a team that understands the failure modes uses the tools more, not less, because they know which situations are safe.

Make sharing routine

The fastest way for a team to get good is for one person's discovery to reach everyone.

A shared document of prompts that worked, or five minutes in an existing meeting. Not a new ceremony — the ceremony will die. Attach it to something that already happens.

Make it safe to say it went wrong

If admitting AI produced something bad is embarrassing, people will hide it, and you will find out from a client.

Say plainly, once, that the tools get things wrong, that catching it is the job working correctly, and that nobody is in trouble for flagging it. Then behave that way the first time it happens, because that is when everyone is watching.

Expect uneven adoption

Some people will run with it. Some will try once and stop. Some will not engage at all, and some of those are right — not every role has an obvious use.

Chasing universal adoption wastes goodwill. Support the people who are interested, make the boundaries clear for everyone, and let it spread by demonstration.