I wrote the soft version. I was wrong.
A while back I published a piece telling business owners to stop selling AI to their employees. Resistance is fear, not stubbornness, I said. Mandates produce compliance theater. Run a two-week pilot on one task. Make the skeptics your evaluators. Credit the people, not the tool. Give it time.
Every sentence of that was reasonable.
Most of it I’d still tell you to do. Pilot one task. Set standards and hand over examples. Credit the people. What I got wrong was the clock and the consequence: no deadline, and nothing on the line for the people who declined.
Reasonable is the problem. That advice was written for a company whose competition was the firm across town, staffed the same way, moving at the same speed, carrying the same payroll. Against that competitor, patience costs you nothing. You can spend a year building buy-in because they’re spending a year too.
That competitor is no longer the one you need to worry about.
Your competitor has nobody to convince
The startup forming right now to take your customers has a slogan on its wall. AI first. You’ve seen it on a hundred landing pages and filed it under marketing.
It isn’t marketing. It’s an org chart.
Picture that company from the inside. Nobody there has a job that predates AI. Nobody has a workflow they’re attached to. Nobody has to be persuaded, piloted or credited into using the tools, because the tools were there before the people were. The founders didn’t adopt AI. They built on top of it, the way your company was built on top of email, and nobody at your company ever ran a two-week pilot of email.
This isn’t a thought experiment. A media company I work with runs its entire writing operation with three people, and it sits at number two in its market. An insurance company handed the dialing to software, moved its cold callers onto closing, and closes more deals now than the full floor ever did. A contractor spends almost no time on proposals and paperwork. None of them hired their way there. They reassigned.
When three people can produce what your forty produce, they don’t need to beat you on quality. They beat you on price, and you never see the deal. The customer just stops calling.
The speed is the part nobody prices correctly. That startup isn’t growing at the rate you grew. It’s growing at the rate the models improve, and the models improve every few months. Every quarter you spend building consensus, their cost advantage widens on its own, without anyone there lifting a finger.
Humans with tools win. Every time.
Take the letters AI out and this is the oldest pattern in commerce. Put two groups of people in the same market, hand one of them the new tool, and wait. The group with the tool wins. Not usually. Not in most cases. Every time, given enough time.
The contractor who bought an excavator finished in a day what a crew with shovels finished in a week, and the crew with shovels stopped getting the job. The accounting firm that moved its ledgers onto a spreadsheet closed the books in an afternoon and priced the firm still working in pencil out of the market. The travel agency with a storefront and a phone lost to a website that never closed. Then software did it to every industry at once, and the firms that filed it under IT expenses instead of under the business itself are the names you no longer hear.
In none of those cases did the people without the tool get a vote. They were not out-argued or out-managed. They were out-produced, and the customer moved to the lower price without ever hearing the debate.
Every one of those shifts took a decade or more to run its course, which is why the losers had time to tell themselves it wasn’t happening. AI is the first that doesn’t extend that courtesy. The tool improves on a schedule of months, so the gap between the people using it and the people refusing it isn’t fixed. It’s widening while you read this.
The arithmetic of waiting
Here is the choice as it actually stands, with the language of culture and buy-in stripped off.
Option one: a few people who will not use AI leave the company this quarter. It’s painful, it’s personal, and some knowledge walks out the door with them.
Option two: everyone leaves in two years, because there is no company.
That is a schedule, not a threat. Retooling a business around AI takes a year of rebuilding workflows, taking wrong turns, retraining people and finding out which tasks the tools handle and which they don’t. Budget a year. The startup finished that year before it hired its first employee. You haven’t started it.
Begin today and you’re a year behind. Begin next quarter and you’re a year and a quarter behind, and the gap compounds, because the thing you’re chasing moves too.
So the question was never whether a few departures are worth it. The question is which layoff you’d rather explain: the small one you chose, or the large one you didn’t.
The new rule: everyone, with a date
Every employee uses AI in their work. Not the curious ones. Not the departments where the fit is obvious. Everyone, from the front desk to the CFO, and that includes you.
Put the standard in writing. Set a date, and ninety days is generous. Pay for the tools, pay for the training, assign someone to help, and make it plain that the help is real and so is the deadline.
Then hold the line. Someone who won’t meet the standard by the date is told so, given notice, and replaced by someone who will. Treat it the way you already treat any other performance standard in the company, because that is what it is now. Showing up and refusing to use the tools the job requires is not a personality trait. It’s a decision not to do the job.
Before you do, say the part that’s yours out loud. If your people treat AI as optional, it’s because you told them it was. For a long time the message from the top was try it if you like, no pressure, and they heard it. The standard is new because you’re only now setting it. That doesn’t change what happens on the date. It changes how you deliver it.
It’s harsh. I’m not going to dress it up. Measure it against the alternative and it’s still the kinder option by a wide margin: a handful of people with notice and a reference now, or the whole payroll with neither later.
One distinction matters here, and it’s the one the soft version blurs. Can’t and won’t are different problems. The employee who is struggling to learn gets time, training and a patient manager, and most of the people you’re worried about are this kind. They’ll surprise you once it’s clear the question is no longer optional. The employee who has decided not to learn gets the date. That person you cannot carry, however good they were at the old version of the job.
What to start today
Today means today, before you close the laptop, not after the offsite.
Think in tasks, not titles. Before you post the next opening for an editor or a marketer, break the role into the activities it actually performs: the research, the first draft, the revisions, the scheduling, the reporting. Ask which of those an AI handles now, and wire the ones it can into a process that delivers the result. Some roles turn out to be three tasks and a judgment call. Hire for the judgment call.
Build one thing this weekend. Pick one real task from your own week and automate it, badly, over a weekend. You’ll learn more in those two days than in six months of reading about AI, and the hours go in once and pay back every week after. If you own the company, this one isn’t optional. Nobody below you will take the standard seriously until you’ve met it yourself.
Set the tools up like you’d onboard a new hire. Vague instructions produce generic work, from people and from models alike. Define what good looks like, hand over examples of it, spell out the rules, and give feedback on the output until it’s right. The companies getting generic results from AI are the ones that gave it generic instructions.
Write the standard down before you leave the office. One page. What everyone is expected to use, by when, what help is available, and what happens if they don’t. Date it. Everything else on this list is easier once that page exists.
Give them a fifth of the week
Of everything above, the one I’d fight hardest to keep is this: every employee gets 20% of their week, paid, to automate their own job.
The person doing a job is the only one who knows where the repetitive parts are. Which form gets filled in twice. Which report nobody reads. Which spreadsheet is rebuilt by hand every Monday morning. A consultant can’t find those. An innovation team can’t find those. The bookkeeper finds them in a week, if given the week.
So give them the week. One day in five, the tools bought, someone to call when they’re stuck. What comes back isn’t a tidier job. It’s a department’s output from one desk.
Here is the part that should change how you hire. The employees who take this seriously are turning into something that didn’t exist three years ago. They’ll arrive at their next job with their own agents, their own workflows and their own tools, and do on day one what used to take a team. Those are the people you want, and they’ll take the offer from the company that gives them room to build. Right now that company is the startup. It doesn’t have to be.
If you’re the one refusing
This article will get forwarded, so let me say it to you directly.
Your boss is about to read this and set a date. Set your own first. Pick the ugliest task in your week and spend a weekend making a machine do it. You’ll hate the first two hours. By Sunday night you’ll have something that works, and a different conversation on Monday.
Nobody got a vote on whether the tools changed. You get one on which side of the date you’re standing on.
The risk I had backwards
I used to think the risk was moving too fast and losing good people.
The risk is moving too slow and losing all of them.
The company that will replace you has already made this decision. It made it on the day it was founded, and it never had to give anyone notice to get there. You do. That is the one disadvantage you carry that can be gone by the end of the week.
