The Six Percent That Brought the Engineers Back

Standard Chartered says AI will replace “lower-value human capital.” Ford, IBM, and Klarna already ran that experiment. The results came in this summer.

On May 19, Standard Chartered CEO Bill Winters told investors in Hong Kong that the bank would cut more than 15 percent of its corporate function roles by 2030, roughly 7,000 jobs. He was careful to say it was not cost-cutting. It was, in his words, replacing “lower-value human capital” with investment in technology. The share price went up.

The phrase traveled fast, and it deserved to. But the more interesting story that season came from companies that had already run the experiment Winters was describing, and were now walking it back.

Ford has spent three years bringing back about 350 veteran engineers to fix vehicle quality problems its automated systems missed. Charles Poon, Ford’s VP of vehicle hardware engineering, told CNBC the company had assumed that feeding AI its design requirements would produce a high-quality product without experienced humans overseeing the work. It did not. The returning engineers rebuilt the AI tools and now lead the troubleshooting sessions, and CEO Jim Farley has credited the effort with hundreds of millions of dollars in savings.

IBM’s version is more precise, and more instructive. The company deployed an AI system to handle human resources requests. It resolved about 94 percent of them. The remaining 6 percent, which included cases requiring ethical judgment, turned out to be the ones that mattered, and there was nobody left to catch them. IBM has since announced it will triple its US entry-level hiring across every business unit in 2026. Its chief human resources officer, Nickle LaMoreaux, put the reasoning bluntly at a Charter AI summit: stop hiring at the bottom and within a few years there is no pipeline at all.

Klarna is the case everyone cites. It replaced roughly 700 customer service staff with a chatbot, cut headcount sharply, watched customer satisfaction slide on the complicated and emotional cases, and reversed course. CEO Sebastian Siemiatkowski now tells the market that customers need to know a human will always be available. The company reported reaching break-even in May, after it started hiring people again.

The pattern behind the anecdotes

Forrester’s 2026 Future of Work report found that 55 percent of employers regret AI-related layoffs. Robert Half says 32 percent of US hiring managers who cut roles for AI are already rehiring for the same or similar positions. Gartner projects that by 2027, half the companies that cut customer service jobs for AI will hire back into equivalent roles. Metaintro, a job search platform, has tracked a surge in reposted junior positions that closely resemble jobs eliminated six to twelve months earlier.

Sam Altman has said some of this is “AI washing,” companies attributing budget-driven cuts to AI because it makes a better story for investors. That is surely true in places. It does not explain Ford, which was not short of money, or IBM, which built the system, ran it, measured it, and hired anyway.

The 94 percent is the cheap part

There is a reading of the IBM number that counts as a win. Ninety-four percent automation of a corporate function is remarkable. But look at where the residual sits. The 6 percent was made up of the ambiguous cases, the ones with a policy conflict or a person in distress or a situation nobody had written a rule for. Those are the cases that decide whether an HR function is trusted. They were also the cases the people being replaced had spent years learning to handle.

The same shape shows up at Ford. Automated quality systems catch what they were trained to catch. The defects that reach customers are, almost by definition, the ones outside the training set. Ford did not bring 350 engineers back to do the 94 percent. Their job is spotting the 6 percent before it ships.

That is the flaw in “lower-value human capital” as a way of thinking about a workforce. Value in a knowledge role does not spread evenly across the tasks. It concentrates in the exceptions,

and the exceptions are invisible in a headcount spreadsheet because nobody logs the disaster a senior person prevented on a Tuesday afternoon. Companies cut the role based on the 94 percent and feel the loss on the 6.

“Companies cut the role based on the 94 percent and feel the loss on the 6.”

 

What a reversal actually costs

Time magazine described the emerging cycle: announce the AI, downsize, wait six to twelve months, discover the system handles around 60 percent of the job, rehire. The rehire is rarely a clean restoration. Ford’s returning engineers came back to overhaul the AI, not to work around it. IBM is rebuilding the junior pipeline that feeds its senior bench. Klarna’s new model keeps the bot for volume and staffs humans for judgment.

In every case the company ended up with the hybrid it could have designed in the first place, minus the severance, the rehiring premium, the lost institutional knowledge, and a year of customers or employees living with the gap.

The question Winters’s framing skips is the one every leader should be asking before the cut rather than after it: which parts of this function are the 94, and which are the 6? You cannot read that off an org chart. You find it by mapping the process at the level of individual decisions, running the AI against it, and watching where it fails. That is unglamorous work. It is also far cheaper than the route Ford took.