They fired the humans, bought the AI, and are quietly hiring the humans back

For two years the announcements came thick and fast: support teams replaced by chatbots, content teams replaced by generators, engineering headcount "optimised" with a straight face. The share prices nodded along. And now, quietly, in the way large organisations do everything embarrassing, the humans are coming back.

The receipts

Klarna is the tidy case study: roughly 700 customer service roles handed to an AI assistant, followed by falling satisfaction scores, followed by the CEO's admission that "we focused too much on efficiency and cost. The result was lower quality", followed by rehiring. Ford brought veteran engineers back after quality recalls climbed. And the pattern is now measurable rather than anecdotal: research by Orgvue found 55% of leaders who made AI-driven redundancies later admitted they were a mistake, and a 2026 survey found two-thirds of the companies that cut for AI have since started rehiring. My favourite detail belongs to Gartner, who predict that by 2027 half of these organisations will have rehired for the same work "under different job titles". The work came back. The dignity of admitting it did not.

The category error

Here's the thing I've watched from inside enterprise cost programmes: the spreadsheet counts heads, not work. So when AI arrived, the question asked in a hundred boardrooms was "whose job can this replace?" rather than "which work should this remove?". Those are profoundly different questions. The people who were cut turned out to be doing two jobs: the visible one on the org chart, and an invisible one made of context, judgement, exception-handling and trust, which nobody could see until it stopped happening. The chatbot answered the easy tickets. The hard tickets, the angry customers, the ambiguous cases, the things that quietly protect revenue, had nowhere to go.

What this means at your size

You might read all this with some satisfaction and conclude AI was oversold. Wrong lesson. The technology did exactly what it does well; it was pointed at the wrong target. And the right target is far easier to see in a twenty-person business than a twenty-thousand-person one: it's the work that gets between your people and making money. The rekeying between systems. The chasing of information someone already has. The assembling of quotes, reports and follow-ups from scratch every single time. The admin that eats the hours your customers would happily pay for. In most SMEs that's five or more hours per person per week, hiding in plain sight, and none of it is anyone's actual job.

So the rule, learned at someone else's expense: automate between the work, not instead of the people. Give every person their five hours back and you've gained a day a week of capacity without a single goodbye, without the rehiring U-turn, and without your customers ever meeting a machine pretending otherwise. Efficiency isn't fewer people. It's more margin per human hour, and the businesses getting that right are doing it quietly, in the back office, where it compounds.

Where to start

Make the invisible work visible: have each person list what they did last week that a customer would refuse to pay for. That list is your automation map, and it's usually a revelation. Score it honestly for value, effort and risk, keep AI away from anything customer-facing until it has earned its place, and give every use case a kill test. If that sounds like a fortnight's structured work, it is, and it's precisely what the AI Context Sprint was built to do.