Don't Roll Out AI to Your Whole Team

We often see owners treat AI like a company-wide announcement. Everybody gets the login on Monday, everybody sits through the training on Tuesday, and by Friday everybody is supposed to be working differently.

Well... it almost never goes that way.

Let me give you an example from a client of ours:

They have several people who touch hiring. One handles people on the ground year round. One hires for specific programs. Another looks after the folks already inside the company who fill different roles depending on the season. Each of them does their portion of the job, and each of them does it well.

And still: confusion, conflicting decisions, and the same work getting done twice by two people who didn't know the other one was doing it.

Nobody in that story is bad at their job. The job itself lives across three heads and no single place.

Three things that don't fix it

When an owner sees that mess, there are usually three instincts.

Write a policy. Now you have a document nobody reads and the same three heads.

Add a standing meeting. Now you are paying for the confusion in calendar time, every week, forever.

Hire a coordinator. Now you have added payroll to manage a problem instead of solving it, and the new person has to learn what the other three already know.

So here's the rub: all three of those are attempts to move information between people. That is the actual problem. And it is the kind of problem that is hard to find a good answer for at this size of business.

The bar, not the headcount

Here is what we actually do.

We look for the people you would trust most to explain how something works and why it is done that way. Not the fastest or the loudest. The ones who carry the reasoning.

Then we give them an AI operating system and teach them to run it.

In that hiring team the answer happened to be one person, because one person understood how all three sets of decisions fit together. On a different team it might be two or three, working different parts of the same process.

You are not picking a headcount. You are applying a bar, and you focus first on the people who clear it.

The test isn't the number. It's whether you can point at each name and say which process they would carry and why they are the one who understands it. If you can't do that for someone, they are not a first seat yet.

I wrote in issue 6 about who this person tends to be, and why it is usually not the most technical person on the team. This piece is about something different: how many of them you start with, and what you actually hand them.

Why not everybody at once

I have watched the everybody-at-once approach up close.

Getting a whole team to move at the same relative speed is close to impossible. Some people feel held back waiting for the group. Others feel left behind by it. This leaves a higher chance that both groups check out, and you paid for every hour of it.

It is also worth knowing where the usage actually sits. OpenAI published research this year on how organizations use ChatGPT, drawn from about 1,500 companies and 17 million messages. Early-career workers send meaningfully more messages per week than the average person at their own company, while executives send the fewest of anyone.

The people at the top are talking about AI more than they are using it.

The first seats are the harder job, not the prize

I want to be careful here, because "start with your best people" can sound like your best people get a reward and everybody else gets to watch. That is simply inaccurate.

Somebody has to find out what actually works before the whole company is asked to change how it works. You send the people most likely to come back with a real answer.

These first seats absolutely get benefits, but they also sit with the clunky early version. Their reward is that they are the ones who gain the most momentum in seeing the direct impact of AI on their day-to-day work and results. It's then up to them to explain it to everyone else afterward and answer the hard questions about it, and our job to make sure they can.

It is the harder job, and you hand it to the people who can carry it.

What everyone else gets out of it

The rest of the team is not sitting on the sidelines.

Every one of them holds pieces of information that only they have. The people in the first seats need those pieces to get output worth anything, so a good part of the work is building the connection points between them: what gets asked, what gets handed over, where it lands, who checks it.

But the bigger thing is what your team sees.

There is a wide difference between leadership announcing that the company is doing AI now, and a colleague walking over with something that works and offering to show you how it runs. The first one asks for belief. The second one is proof.

Proof spreads. Announcements just get nodded at.

Why waiting is the expensive choice

From that same OpenAI research: nearly half of the growth in usage came from companies that had already started, going deeper, not from new companies signing up.

It's the difference between a company that gains the skill of using AI versus those that just have a login.

That gap compounds. And it compounds harder in a local market than it does at enterprise scale, because you are competing against a shorter list of businesses and the customer can feel the difference in a week.

Here is what it looks like in a business your size.

You hire for a role your best people could have covered with a system behind them. That is payroll you carry for years. To be clear, I am not talking about cutting anybody. I am talking about the next hire you were about to make, and whether you actually need it.

You add people to manage complexity instead of removing the complexity, and the complexity grows to fit the people.

You can take the momentum now instead at a better cost than in the future.

You may already know who they are. You probably thought of somebody a paragraph ago.

Start there, and let the rest of your team watch what happens.

Execute. Rinse. Repeat.