The customer has already ordered. The order is sitting in an email. The salesperson opens the email, copies the order into the company’s system and passes it on. For that part of the day, they might as well be an invoicing clerk.
I co-own several companies. I organize the work, oversee it and sometimes do it myself, so I see how people spend their day, including the salespeople I work with. Someone had to pass that order on, or the customer might never get the goods. I’ve never found a good answer to why the company needed a person to hand it information it already had. When AI takes over that job, AI gets the credit. Some of it belongs to whoever finally connected the inbox to the order system. My suspicion is that a lot of the work we now say AI is taking away is work we could have organized out of existence years ago.
There was another job I wanted salespeople to do. After a call, somebody had to record who they’d spoken to, what they’d promised and when they’d call back. If I took over the account the following week, I needed to know all three. Plenty of salespeople skipped the entry in the CRM, and whoever called the customer next was working blind. I understood them. They’d had the conversation once. Now they had to sit down and tell it all over again to the computer.
Today AI can draft the entry. The salesperson checks what it says about the promise and the date, then clicks to save it. With the emailed order, we’ve saved somebody the trouble of entering the same information twice. The call is different. Quite often we’ve finally got a piece of work done that used to get skipped. The next person who picks up the phone has some idea what the customer is waiting for.
Take an oil change. The mechanic writes down how much oil went into the car and half an hour of labor. The office types those figures into the invoice. Even on a day when everybody gets every figure right, two people have spent time entering the same information. And each time a number changes hands, somebody has another chance to get it wrong. Suppose the mechanic writes 0.5 hours and the office enters 5. Catch it before the invoice goes out and the office has to dig out the slip, check it and change the invoice. Catch it next month and accounting has to go back to the workshop to find out how long that oil change actually took.
Now let the mechanic select the oil and the hours on a screen. The customer gives a tax ID, the software pulls in the company details, and the same figures go straight onto the invoice. The mechanic can still make a mistake. But if the hours are right on the screen, there’s no second pair of hands to turn half an hour into five. That saves a few minutes of typing on every invoice. On the one that would have gone wrong, it may also save an accountant a morning spent chasing an error the workshop itself created.
Ordinary software could do this years ago. AI makes it easier for a small workshop to build the connection. On the websites I work on, I can describe a change in my own words, let AI write the code and check the result myself. AI’s real contribution here is that it makes the reorganizing cheap. I wonder how many other jobs we’re about to credit it with eliminating could have disappeared long ago, if somebody had got around to changing the way the work was done.
Michael Hammer, the MIT computer scientist who became the best-known advocate of “reengineering”, described much the same problem in 1990, in an article called Reengineering Work: Don’t Automate, Obliterate. At Ford, accounts payable had to reconcile a purchase order, a record of what had arrived and the supplier’s invoice. Ford’s accounts payable department in North America employed more than 500 people. At Mazda, a smaller company, five people handled accounts payable. Ford changed the procedure. When the goods arrived and matched the order, the warehouse recorded the delivery and that triggered payment. Ford asked suppliers to stop sending invoices. Hammer reported that where Ford adopted the new process, it needed 75 percent fewer people in accounts payable. Nobody automated the invoice. Ford stopped asking for it.
Hammer saw this in 1990, at one of the best-known companies in the world. Thirty-six years later, plenty of offices still retype what they’ve already received. A manager who needs an invoice out today will ask somebody to type it up. It takes a minute. Changing how the office works takes time, money and someone willing to answer for it if it goes badly. On any given Tuesday afternoon, the minute wins. The trouble is that it can win every Tuesday for twenty years.
Hammer’s 75 percent were people, too. In the 1990s ‘reengineering’ became a polite word for layoffs. When we talk about AI taking jobs, we are often talking about the same thing arriving thirty years late.
People are already losing their jobs because of AI. It happened at one of my own companies. Microsoft is a harder case. When it announced about 4,800 job cuts in July 2026, the headline from Reuters, via KSL, placed Microsoft in an AI-driven wave of tech layoffs. Microsoft’s memo talked about changes in how it worked with customers and a restructuring at Xbox. It also said the eliminated roles “are not being replaced by AI.” How many of those 4,800 people lost their jobs because a machine could now do their work? I don’t know. I know what happened in our studio.
We once employed nearly thirty 3D artists, creative, intelligent people, some trained in architecture, others at art school. My estimate is that 90 percent of their time went into reproducing things. Give someone a bottle and they could spend hours recreating the very bottle you’d just handed them. You could scan it, if the scan worked, and then an artist would have to clean up the model anyway. Sometimes we had people making versions we knew the client wouldn’t like because we had to show several options.
AI changed what a handful of people could do in that studio, and we could produce better work faster. We still needed artists to spot what was wrong and get an image into a state we could send to a client, but we needed far fewer hours from them. Our clients knew it, too, and they pressed us to lower our prices. We still had clients and orders. A few people could now do the work that had kept a team of nearly thirty busy.
There’s a second problem with the story about freeing people from dull work. Jaspreet Ranjit and colleagues looked at which tasks AI could take on or speed up, and which parts of their work people found satisfying. They found that some of the tasks particularly exposed to AI were ones people enjoyed or that gave them a sense of control over their work. I think of an artist who would gladly let a machine clean up the scan of the bottle. The part the client hands to AI may be the part the artist loved doing.
Suppose this reaches a third of the people who work for a living. I don’t know if it will. I do know how easy it is to say that people will retrain. Some will. Some will look for another job and never find one. What would I tell them about all the years when the work they did suited the rest of us just fine?
I remember standing in front of our artists. My business partner and I had to tell them we had no more work for most of them. We had to finish the last jobs, make sure everybody got paid and say goodbye. It was awful. I still think we’d be better off if fewer people spent their days copying orders, rebuilding objects that already existed and fixing mistakes we could have prevented. But when we talk about who’s responsible for what happens to those people, I think about who kept asking them to do that work, year after year. In our studio, I was one of the people asking.
