Everyone Is Adding A.I. Almost No One Is Subtracting.

Companies keep announcing what they will add. The harder question, and the more valuable one, is what they will stop. 

The software upgrade went live overnight, and it worked. 

Most of the team that made it work left the office at five in the morning. At 9:01 that same morning, their boss stood at the door and yelled at everyone who arrived after the hour. 

I have thought about that morning for years, and less about the yelling than about the arithmetic behind it. In that boss’s accounting, the upgrade was a gain and the night behind it cost nothing. 

She was an extreme case. The accounting is not. Most leaders who pile new work onto a team are not cruel. They are keeping books with a single column, the one that lists what they are adding. And right now, in company after company, what is being added is artificial intelligence. 

Money is not what is holding it back. In January, Boston Consulting Group reported that 94 percent of the 640 chief executives it surveyed would keep investing in A.I. at current or higher levels even if it did not pay off within a year. Nor is adoption the problem. A survey of 150 North American C-suite leaders published in June by the University of Phoenix found that 63 percent had deployed at least one A.I. use case, but fewer than one in three were using it to change how work gets done. 

What is left is the part that never appears in a budget. Writer, an A.I. software company, surveyed 2,400 employees and executives this year and found that 79 percent of organizations face challenges adopting the technology, even though 59 percent are investing more than $1 million a year in it. In McKinsey’s 2026 survey of 1,719 respondents, 80 percent said A.I. had improved their own productivity, yet only 37 percent attributed any earnings impact to it, about the same share as a year earlier. The technology is working for individuals. It is not yet working for the organizations they belong to. 

I cannot prove the missing ingredient is subtraction. What I can offer is a pattern. I have been fortunate to spend almost 30 years leading enterprise technology transformations in pharmaceutical, healthcare and consumer goods companies, and I have watched it repeat through every wave. The constraint is rarely technology and rarely money. It is capacity: the finite supply of attention, sponsor time and change that people can absorb at once. Add without subtracting, and you do not get transformation. You get exhaustion with a project code. 

The strain does not fall evenly. McKinsey found that 47 percent of mid-level managers and individual contributors reported at least one negative effect of A.I., compared with 31 percent of executives and senior managers. That is what a one-column ledger looks like from the middle of the organization, I would argue: the people who feel the pile first are the people farthest from the decision to add to it. 

So where does subtraction start? With the projects that stay alive because ending them would be awkward. Every organization has them. They are not free; they draw on the same small well of talent and sponsor attention that the A.I. agenda needs. The test is plain: Would we fund this today, knowing what we know now? 

It also starts with the process that A.I. is being laid on top of. Ask why a step exists, and the answer is often a shrug: something broke years ago, nobody had time to fix it, and the workaround became the process. Point A.I. at that step, and you have automated a scar. Consider the weekly status report, compiled by hand, reviewed by a manager, summarized for a director and reformatted for an executive. A.I. can draft the first version in minutes, and the temptation is to celebrate the speed. The better question is which of those hand-offs should exist at all. McKinsey found that nearly three-quarters of its high performers had fundamentally redesigned workflows because A.I., compared with about one-quarter of other respondents. In my experience, redesign is mostly a decision about what to remove. 

There is a fair objection. A.I. is supposed to be the subtraction: it removes drudgery and hands people time back. Sometimes it does. But the time returned is not the same as time reclaimed. If the hours saved are filled by the next initiative the same afternoon, nothing has been subtracted; the pile has simply been refilled. A second objection is that no company can afford to slow down while competitors speed up. I agree, which is why I am arguing for choosing, not for pausing. 

Choosing is hard because it is personal. Behind every project is someone’s credibility, someone’s team, someone’s sense of purpose. End the work carelessly and you end that person’s trust in you along with it. The leaders who do this well say aloud what the project achieved, make sure its owners feel heard, and help them land somewhere new with a real plan rather than a vague promise. Transformations fail because of people, not systems, and subtraction is where that is hardest to avoid. 

Here is the rule I would give any leader. For every new A.I. initiative, write down what it replaces, who owns the wind-down and when. If you cannot fill in all three blanks, you are not launching a strategy. You are adding a pile. 

The upgrade, as I said, worked. The ledger that counted it had no column for the night.