How to get a whole team using AI, not just the curious two

Every company I talk to has the same two people. They've built their own prompts, they're quicker than everyone else, and they can't understand why nobody copies them.

Nobody copies them because what they've built only works on their laptop, with their accounts, the way they think. Getting from those two people to the whole team is not a motivation problem. It's a setup problem, a sharing problem and a trust problem.

Here's what worked at a 25-person marketing agency, where 22 teammates are now set up and the team shares 16 processes between them.

1. Make setup take one command

At the agency, the "team" AI workspace was really one person's personal setup. Teammates on Windows couldn't get it running at all. Every failed install produces someone who has decided AI isn't for them.

So setup became one command that takes a fresh laptop, Mac or Windows, to a working state. A health check tells you what's wrong in plain words. Updates are one command too. If getting started takes more than a few minutes, most of your team will never get started.

2. Keep secrets out of the chat

People paste passwords and API keys into chat windows because nobody gave them another way. Give them another way. A small, safe step for saving a key, so it never appears in a conversation, removes a real risk and a reason for careful people to stay away.

3. Make sharing safe for people who aren't technical

The moment that matters is when someone builds something useful and wants the rest of the team to have it. If that needs a developer, it won't happen.

At the agency, a teammate says "share this with the team." The change is checked automatically, first by fixed checks and then by an AI review. Safe changes merge on their own. Anything risky waits for a person. At least five teammates besides the founder have now contributed their own.

4. Train by team, then keep showing up

One big kickoff doesn't work. What worked was an onboarding session for each team, then a weekly check-in, one-to-one help for anyone whose setup misbehaves, and written best practices people can go back to.

Adoption isn't an event. The first fully automatic share-and-merge at the agency came about ten weeks in. Most of the value arrived after the point where a typical rollout has already been declared finished.

5. Let the team own it

Most of the agency's shared processes were written by its founder and team, not by me. That's how it should be. They know the work. My job was the plumbing and the training that made it possible for them to build their own, and to trust what their colleagues built.

If every new process has to come from a consultant or from IT, you'll get a handful. If anyone on the team can add one safely, you get a library that grows on its own.

Prune as you go

Shared workspaces collect junk. Part of the work at the agency was slimming the workspace from 2,282 files to 414. Less to read, less to break, and easier for a new teammate to see what's there. A smaller shared library that everyone understands beats a huge one nobody trusts.