The Most Expensive Way to Use AI

The most expensive way to use AI is to fire somebody for it. That sounds backwards, so here’s the math a lot of companies are quietly running right now and hating the answer to. You let go of a person earning, say, 55,000 dollars, because the AI can handle most of what they did. It feels like a saving, and for about six months it is. Then the bill arrives. The AI did about 60% of that job, the repeatable part. It could not do the other 40%: the judgment calls, the weird exceptions, the customer who needed a human. So you go to hire that back. Except now you’re not hiring a 55,000 dollar person. You’re hiring someone who can do the hard 40% and manage the AI on top of it, and that person costs 75,000. Add recruiting, training, and everything the person who left knew that nobody wrote down. The saving is gone. You’re poorer, and you’re back where you started. So this isn’t about whether to use AI. Of course you should. I use it every single day and it’s the best money I spend. This is about where to point it, so you get all of the upside and none of that boomerang. And it comes down to one shift in how you look at your own business. You stop thinking about replacing people, and you start thinking about replacing tasks. The method has four steps, and you can run it on your own team this week. Step 1: Audit the task, not the job Don’t look at a role and ask, can AI do this job. That’s the question that gets people fired and then rehired. Instead, take one person and write down what they actually do in a week. Not their title. Their tasks. Most people, when you really list it out, are doing something like 15 to 20 distinct things a week. Get them all on paper. Now draw a line down the middle. On one side, the routine and repeatable work, the stuff that happens the same way every time. On the other, the work that needs judgment, or context, or a real human relationship. Almost every job is a mix of both, and this is where most owners get it wrong: they think they’re paying for the whole list. They’re not. They’re paying for one side of it. Step 2: Understand the 60/40 line This is where the machine actually stops. The routine side, roughly 60% of the work, AI can genuinely take a big bite out of: drafting, sorting, summarizing, answering the same question for the hundredth time. Hand that over gladly. But the judgment side, the other 40%, is where it falls apart. Not because it’s lazy, but because that work needs something the model doesn’t have. What it doesn’t have is your context. A model will write you a confident, competent first draft of almost anything in about four seconds. What it can’t tell you is whether that draft is right for your situation, your customer, this specific moment, because it has never been in your business. The first draft was always the cheap part. Knowing whether it’s right is the expensive part, and that still lives in a person. The working rule for what to hand over: If you can write clear instructions for it, give it to the AI. If it needs someone who knows your world, keep the person, and let the AI take the busywork off their plate. The boomerang, with real numbers Let’s put numbers on that boomerang from the top, because this is the part that should stop you before you cut anyone. Start with the 55,000 dollar salary you think you’re saving. The AI handles the 60%, so call it about 33,000 dollars of actual time saved. Good so far. But the 40% it can’t do still has to get done, so you hire it back, at a higher rate, because now you need judgment plus AI-wrangling. Add recruiting and training and lost knowledge, and you’re out another 30,000. Do the arithmetic. The saving didn’t shrink. It vanished. This isn’t a scary hypothetical. It’s already happening at scale. The staffing firm Robert Half found that nearly a third of companies that cut jobs for AI have already rehired for the same roles. Gartner expects at least half of the customer-service cuts to reverse by 2027. And a separate survey found 55% of executives who replaced people with AI already regret it. The boomerang isn’t a risk. For a lot of them, it’s a receipt. Step 3: Protect the context-holder Even the rehiring doesn’t fix the real problem. When that experienced person walks out the door, some of what they knew walks out with them, and you can’t buy it back at any salary. Every business has that person who’s been there eight years and just knows things. Which client will dispute every single invoice. Which supplier goes sideways every December. Why you stopped doing it the old way, because you tried, and it blew up. None of that is written down anywhere. It was never in a job description. It lives in one human head, built up over years of paying attention. And there’s a lovely twist the panic completely missed. A Harvard economist who studies this, David Deming, found that AI doesn’t lower the value of human judgment and people skills. It raises it. When the routine work gets cheap and instant, the scarce thing, the thing everyone suddenly needs more of, is the person who can tell whether the machine’s answer is any good. So the veteran you were about to replace is the exact person who just became more valuable, not less. The rule in one line If you take one sentence from all of this, take this one. Hand the machine the bounded, repeatable task. Keep a human on anything that needs judgment, real context, or a relationship. Augment your people; don’t try to replace them. That single line would have saved most of the companies in the headlines a fortune and a lot of embarrassment. And the reason it works is the whole point: the part of a job a machine can copy was never the valuable part. The valuable part was always the judgment, the context, and the trust, the things that only sit inside a person. Aim AI at the task instead of the person, and you keep everything that made that person worth paying. You just hand them a very fast assistant. That’s not a threat to your team. It’s a raise in what your team can do. Step 4: The small-business amplifier One more, and this one is only for those of you running something small, because it changes the stakes completely. When Amazon gets one of these calls wrong, it loses a rounding error. When you get it wrong with 15 people, you can lose the whole thing. If a giant sheds 5% of its customers to a bad chatbot, it barely notices. If you lose 5% of yours, it’s probably your five best, the high-touch, high-margin ones who needed a human and didn’t get one. But that same smallness is also your advantage, if you use it. A giant finds out a year later, buried in a quarterly report, that the automation backfired. You find out in about a week, because your customers call you, and sometimes they call you by name. So move slower on cutting and faster on listening. Hand the AI a task, watch what actually happens to your customers, and keep the humans anywhere the feedback gets worse. Your size is a superpower here. Don’t waste it copying a giant’s mistake. The short version Don’t do what the headlines told everyone to do. Don’t look at your team, see salaries, and start subtracting. Look at the work, task by task, and start sorting. Give the machine the boring, repeatable 60%, gladly. Guard the judgment, the context, and the relationships like the assets they are. Do that, and AI becomes the best hire you ever made, instead of the most expensive one you ever fired.