Start small. That is how AI actually gets used at work.

In a previous role, the AI project that stuck was not a big programme with a launch event. It was a small assistant that answered engineers' questions from our own documentation.

It saved people time in the first few weeks. That made the next conversation with leadership easy. Proof first, then the ask. With security signed off and a few simple rules about data, it grew from one team to twenty five.

I have watched the opposite happen too. A licence for everyone, a training session, a big announcement. Three months later, almost nobody is using it. Same technology. Different start.

It is a bit like opening a restaurant. You do not build a hundred seat dining room on day one. You cook one dish really well and let people tell their friends.

What the good starts have in common

• A real annoyance. Something people already moan about, not something that would be nice to have.
• Your own material. Answers grounded in your own documents beat clever answers from the open internet.
• Security in the room early. Bringing them in at the end is how good projects die.
• A number that matters. Hours saved, tickets avoided, time to an answer. Measure it from week one.
• People choosing it. If it is useful, word spreads. If it has to be forced on people, it is probably not useful yet.

The bit nobody puts in the slides

Adoption is a people problem dressed up as a technology problem. The model rarely decides whether something works. Trust does.

One team. One problem. One honest number. Then grow it.