On July 1, CNBC ran a story with a headline that a lot of executives probably didn't want to read over their coffee: employers who laid off workers for AI are reversing their decisions.
The number underneath it is the part that stuck with me. Fifty-five percent of companies that cut staff for AI now say the decision was wrong. That figure shows up in two separate places, Forrester's 2026 work predictions and a workforce study from the analytics firm Orgvue, which is usually a sign it's real and not a fluke of one survey.
Here's the read that most of the coverage missed. This isn't AI failing. It's people pointing it at the wrong target.
They automated the headcount, not the work
There's a quote in Forrester's research that says the whole thing. When you ask the CEOs who announced "we're replacing 20% of our people with AI" whether they actually had a working AI system in place, "nine out of ten times the answer is no."
They fired first and hoped the technology would show up to cover the gap. It mostly didn't.
You can see the pattern in the receipts. Robert Half found that 32% of managers who cut a role for AI later rehired for the same or a similar one, with finance leading at 44%. And rehiring isn't cheap. Roughly one in three companies that brought back more than half their cut staff ended up spending more on restaffing than the layoffs ever saved.
There's even a market signal now. Goldman Sachs found that as of late 2025, stocks drop about 2% on average after a company blames layoffs on AI. That's a flip from the "efficiency" bump those announcements used to get. Investors got wise to the difference between automating and hoping.
Meet the 6%
The clearest example of doing it right is IBM, and it's worth walking through because it's the opposite of a horror story.
IBM's internal HR assistant handles about 94% of routine HR questions. Millions of interactions, and it cut HR operating costs by a real margin. But the other 6% doesn't go to a bot. It routes to a human, on purpose, because that 6% is the stuff with judgment in it. The sensitive case. The exception. The thing where being technically correct and being right are two different answers.
CEO Arvind Krishna has been direct about the headcount piece too. AI replaced "a couple hundred" HR roles, and the money got reinvested into engineers, sales, and marketing. IBM then said it would triple its US entry-level hiring in 2026. That's not a company shrinking. That's a company moving people off the routine and onto the work only people can do.
That 6% is where your business actually lives. It's the billing dispute that's really about a relationship about to walk out the door. It's the customer whose problem doesn't match any of the three buttons in the script. Automate the 94% and your people get their week back for the 6%. Automate the 6% and you just handed your hardest, highest-stakes moments to something that can't read a room.
What happens when you get the line wrong
Look at the ones who reversed.
A large bank cut around 45 customer service roles back in mid-2025, saying its voice bot had dropped call volume. The union pulled the actual numbers, and call volume was going up, not down. The bank was paying overtime and pulling team leads onto the phones to keep up. It admitted the roles were never actually redundant, brought them back, and apologized. That one predates the current wave, but it's the template for it.
Ford went a different route and learned the same lesson. After years of trimming veteran staff as AI took over parts of quality and design review, quality problems stuck around. So Ford brought back around 350 veteran engineers, some of them retirees, and then topped a major 2026 quality study for the first time in over a decade. The AI couldn't hold what those people carried in their heads. Nobody had written it down, because most of the judgment that runs a business never gets written down.
That's the hidden bill. When you delete a person to save a salary, you also delete years of undocumented context about your customers, your edge cases, your history. The rehired workers know it now, which is why a lot of them are coming back at higher pay. You're not just re-recruiting. You're buying back knowledge you gave away for free.
The small-team version of this
I'll be straight about why this topic hits home for us. Kief Studio is two people, me and my wife Meelie. Between us we cover what would normally take a team of ten to fourteen, and we do it by living on the right side of that exact line.
We didn't hire people we couldn't afford and we didn't pretend a bot could be a colleague. We built our own automation, LTFI, and pointed it at the routine work. The daily content pipeline, the server provisioning, the parts of the job that are the same every single time. That's the 94%. It runs so the two of us can spend our hours on the 6% that actually needs a human who understands the client, the risk, and the call that doesn't have a clean answer.
That's the whole move. A small augmented team wins by being deliberate about which work is routine and which work is judgment, and never confusing the two. Not by having the most seats. Not by having the most bots.
Find your 6% before you automate anything
The companies bleeding money this summer didn't automate too much. They automated to delete people instead of to free them. Those are different projects that happen to use the same tools.
So before you point automation at anything, find your own 6%. Which moments in your business genuinely need a person who can weigh a situation, not just match a pattern? Protect those. Then automate the rest so your people actually have the time to do them well.
Most owners already know where their 6% is. They've just never been given permission to keep humans on it and let the machines take everything else.
If you want to talk through where that line sits in your business, that's the kind of thing we do. Subscribe free at kief.studio for the guides and resources, and the first conversation is always on us. No commitment, no pitch about replacing your team. Usually the opposite.