The Wrong Room

An operator I know bought ChatGPT licenses for his whole team in March. Twelve seats. Six weeks later he pulled the usage report. Nine of the twelve people had logged in exactly once. One person used it every day. Two never logged in at all.

He called it a failure. “We tried AI. The team just won’t touch it.”

Same company. Same six weeks. The tool at the front desk that answers the phone booked a little over forty jobs that week with no human picking up. It texts the customer back. It offers two time slots. It writes the appointment onto the calendar. Nobody at that company calls it the AI. They call it the thing that answers the phone.

Two AIs. One company. One got ignored. One got leaned on every single day. The owner is sure the difference is the model, or the training he didn’t do, or his people being stuck in their ways.

Here’s the thing. It’s none of those. Both tools run on the same kind of engine under the hood. The training gap is real but it’s not the story. And his people aren’t stuck. They adopted one of the two AIs completely. The variable that decided which one isn’t capability. It’s location.

Your team didn’t reject the AI. They rejected the trip.

The trip you can’t see

Think about what the chat-window AI actually asks of a person.

Stop what you’re doing. Leave the screen you were on. Open a different tab. Log into a different tool. Re-type the context the model has no idea about, because it can’t see the email you were just writing or the customer record you had open. Wait for an answer. Read it. Copy it. Go back to the tab you left. Paste it in. Fix the parts that don’t fit.

That’s a trip. Eleven steps, give or take, around a forty-second task.

Now the phone-answering AI. The customer calls the number they already have. The tool answers. The job lands on the calendar the team already checks every morning. Nobody went anywhere. The work showed up where the work already lived.

Right? Same underlying technology. One of them costs a trip every time you use it. The other one costs nothing, because it’s already standing in the room.

Your team isn’t lazy and they aren’t scared of AI. They’re doing the math every operator’s team does, fast and without saying it out loud. “Is the detour worth it this time?” Some days, yes. Most days, when they’re moving, the answer is no. So they skip it. Not the AI. The trip.

The Station Plan, and the room you put the AI in

There’s a way to see your business as a working kitchen. We call it the Station Plan. The human stands at the Hub making the calls. The work moves between stations on the Line. And the whole thing is built so the work flows to the person, not the person running to the work.

A good kitchen doesn’t make the chef sprint to the walk-in freezer every time a plate needs butter. The butter is at the station. The prep is within arm’s reach. The chef stays put and the ingredients come to the hand. That’s not laziness. That’s the design. The second you make the chef cross the kitchen for one item, that item stops getting used. They’ll substitute. They’ll skip it. They’ll work around it.

A chat window is the walk-in freezer. It’s a real room with real value inside it. It’s also across the building.

When you bought twelve seats and told the team “the AI is in there, go use it,” you built a station nobody can reach without leaving their post. You didn’t deploy AI into the kitchen. You parked it in a separate room and hung a sign on the door.

The phone-answering tool got it right by accident. It didn’t ask anyone to come to it. It installed itself at a station the team already worked. Does that make sense? The kitchen never noticed a new station got added, because the new station showed up exactly where a person was already standing.

That’s the whole difference. Not the model. The room.

”But the chat window can do anything”

Here’s the objection, and it’s a fair one. The chat window is flexible. It’ll do a hundred different tasks. The phone tool does exactly one. Why would you trade a tool that does everything for a tool that does one thing?

Because a tool that does everything, that nobody opens, does nothing.

That’s not a clever line. It’s just the arithmetic. Flexibility you have to travel to reach loses, every time, to a narrow tool that’s already in the room. The owner with twelve seats has a tool that can do a hundred tasks and is doing roughly zero of them. The phone tool does one task and does it forty times a week. One of those is an AI deployment. The other is a subscription.

The operator instinct here is to buy range. Get the powerful general tool, and surely the team will find all the uses. They won’t. Not because they’re incapable. Because every one of those hundred uses costs a trip, and a busy person doesn’t take a hundred trips a day. They take zero and get back to work.

So the move isn’t to buy a narrower tool. The move is to stop counting on the team to travel. Take the three or four tasks that actually matter, and get those landing in the room. The general tool can still sit in its room for the person who wants to go visit it. Fine. But your deployment is the handful of jobs you walked to the team, not the hundred you left in the freezer.

”The model didn’t change”

Here’s the thing the AI industry keeps proving and operators keep missing.

There’s a well-known case where a team’s AI adoption sat around twenty percent for months. Same model, same capability, same people. Then someone moved the output. Instead of living in a chat window people had to visit, the AI’s work started appearing inside the tools the team already had open all day. Adoption went to a hundred percent. The model didn’t get smarter. The model didn’t change at all. The room changed. Yeah.

Sit with that, because it rearranges what “an AI strategy” even means.

Most operators think the AI question is which tool do I buy. That’s the wrong question, or at least it’s the third question. The first question is where does the output need to land so a working person never has to go get it. The CRM they live in. The inbox. The scheduling board. The invoicing screen. The job already open on the monitor.

You don’t have an AI capability problem. The capability has been good enough for two years. You have a delivery problem. The smartest model on earth, parked in a room your team doesn’t walk into, gets used exactly as much as a model that doesn’t exist.

What you’ve been measuring

Here’s the thing about that usage report. It fooled the owner, and it’ll fool you the same way.

He measured logins. Chat-window visits. “Nine of twelve logged in once.” He read that as the team rejected AI, and the deeper read underneath it was AI doesn’t work for a shop like mine. Both wrong. The login count measured one thing only. It measured how many people were willing to take the trip. It told him nothing about whether AI was working in his business, because the AI that was working in his business never generated a login. Not one. The login report literally could not see it. It just answered the phone.

I mean. If you’re measuring AI adoption by how often people open the AI, you’re measuring the freezer door. You’re counting trips. The number you actually want is invisible to that report. How much of the real work is now getting AI help without anyone making a trip to get it.

Different question. Different number. The first one tells you whether your team likes detours. The second one tells you whether you have AI in your business at all.

The Monday Move

Pick one task your team is supposed to be using the chat window for. One. The follow-up emails. The quote summaries. The call notes. Whichever one you’ve been quietly annoyed nobody’s adopting.

Now answer one question about it. What tool does the person doing that task already have open when they do it?

Not the AI tool. The other one. The inbox. The CRM. The scheduling screen. The place the work already lives.

That’s your real deployment target. The job this week isn’t get the team to use the chat window more. It’s get that one AI output to show up inside that one tool they already have open. Sometimes that’s a built-in feature you haven’t switched on. Sometimes it’s a small connection between two tools. Sometimes it’s as plain as changing where a draft gets dropped, so it’s waiting in the inbox instead of sitting in a tab nobody opens.

One task. One tool. Move the output into the room.

Then check adoption again in two weeks. Don’t count logins. You won’t see a thing in the login report, and that’s the point. Watch whether the task is getting done faster. Watch whether anyone complains when you take it away. That complaint is the only adoption metric that was ever real.

So.

You didn’t buy the wrong AI. You put a fine AI in the wrong room.

The team that ignored the chat window is the same team that leans on the phone tool without thinking about it. Same people. Same week. The thing that changed wasn’t their willingness. It was the walk.

So before you renew the licenses, or run another training, or decide your shop just isn’t an AI shop. Find out where your people are actually standing. Then bring the AI to them.

Your team didn’t reject the AI. They rejected the trip.

Source influences: Ethan Mollick on the interface disappearing and AI adoption climbing once output landed inside the tools people already used; Andrej Karpathy on the exposure gap. Distilled and operator-translated.

Framework spine: The Station Plan (work flows to the human at the Hub; the human doesn’t run to the work). Read the full framework.