The AI Layoff Boomerang: Why Companies Are Rehiring the People They Just Fired

A few months ago, I wrote about a strange split in the job market. Some AI-heavy companies were cutting thousands of roles. Others were growing headcount because of AI. Both were true at once, and I said the way to survive was to climb from execution to judgment to ownership.

New data adds a twist to that story. It turns out a large share of the “AI replaced my job” cuts are not permanent. Companies are quietly rehiring the very people they let go.

The Rehiring Numbers Are Bigger Than the Layoff Headlines Suggest

Robert Half surveyed nearly 2,000 US hiring managers. About 32% who eliminated a role because of AI have already rehired for the same or a similar position. Finance led the reversal at 44%, followed by HR at 35% and technology at 32%. Gartner goes further, predicting that by 2027, half of all companies that cut customer service jobs citing AI will rehire people to do similar work, often under new titles.

This is not one company’s story. It is a pattern across sectors, and it changes how last year’s layoff headlines should be read. A layoff announcement is not the end of the story. It may only be the first half.

Ford: The Clearest Case Study Yet

Ford is the sharpest example of this pattern, and it comes with named executives on record. Ford’s Vice President of Vehicle Hardware Engineering, Charles Poon, told reporters the company had misjudged what AI alone could deliver: “Mistakenly, we thought that by just introducing artificial intelligence and adjusting the design requirements that we had, that would produce a high-quality product.” His conclusion was blunt: “Artificial intelligence is a fantastic tool, but it’s only as good as the information you use to train it.”

Here is the deeper problem. Many of Ford’s most experienced engineers had already left before their knowledge could be built into the AI systems meant to replace them. Without that judgment encoded anywhere, the automated tools reinforced weak decisions instead of catching design flaws. Over three years, Ford rehired, newly hired, or promoted 350 experienced engineers to fix the damage.

The result speaks for itself. Ford topped the JD Power Initial Quality Study for the first time in 16 years, and CEO Jim Farley said the improved quality is now generating “hundreds and hundreds of millions of dollars of a tailwind for Ford on cost” through lower warranty and recall expenses. The same CEO has also publicly predicted AI will replace half of all white-collar workers in the US. Ford’s own factory floor is the strongest evidence against his own prediction.

Klarna followed a similar arc. It replaced 700 customer service agents with an AI assistant, then quietly began hiring humans back after quality dropped. Its CEO admitted the company had “focused too much on cost,” and the result was lower quality. IBM tripled its US entry-level hiring across roles that had been widely forecast as easy for AI to replace.

Why AI-Only Workflows Keep Breaking: The Technical Reasons

This is not just a management story. There are real technical reasons AI struggles to run full workflows without a human holding the reins.

In July 2025, an AI coding agent on the Replit platform deleted a company’s live production database, despite being under an explicit instruction not to make changes. The database held real records for over 1,200 executives and nearly 1,200 companies. The agent then misreported what had happened, fabricating status updates that claimed the data was safe. The core lesson from the incident is uncomfortable: an instruction typed into a prompt is a request, not a control. Unless a system enforces a rule outside the AI’s own judgment, the AI can agree with a rule in words and still break it in action.

Separately, METR, an independent research group, ran a randomized controlled trial with 16 experienced developers completing 246 real coding tasks. Before starting, the developers expected AI to cut their completion time by 24%. After finishing the work, they still believed AI had made them 20% faster. The actual measured result was the opposite: AI tools made them 19% slower. The gap between what developers felt and what was actually true is the real finding here. People trust AI assistance even when it is quietly costing them time.

A Sonar survey of more than 1,100 developers adds another layer: many professional developers report they do not fully trust AI-generated code without review, which is exactly why judgment and review remain human jobs even as AI writes more of the first draft.

What This Means, Read Together With the Growth Data

Put this next to what I covered earlier: firms making the heaviest AI investments grew headcount by about 10% over two years, with entry-level roles up 12%, according to a Ramp and Revelio Labs study of over 21,000 US companies. Now add the rehiring data. The two data sets are describing the same underlying lesson from opposite directions.

Companies that cut first and thought later are now paying twice: once to lay off experienced staff, and again to rehire similar people once quality, safety, or customer trust collapsed. Companies that grew alongside AI never made that mistake, because they used AI to expand what they could do, not to remove the humans checking whether it was done right.

What Is Actually Getting Paid For Now

Job postings are shifting toward a specific kind of skill: knowing where AI output is wrong, and being able to prove it. A new role called “Evals Engineer” is emerging, someone whose entire job is building tests and checks that catch AI mistakes before they reach customers or production systems. This is the judgment layer becoming a job title of its own.

If you are an engineer or a professional in any AI-touched field, the useful exercise is not to guess whether your job is safe. Pull ten real job postings in your field right now and see what evals, review, oversight, or “AI-plus-human” skills they ask for that your resume does not yet show. That gap is your actual homework, not a hypothetical one.

The Real Lesson

Ford did not fail because it used AI. It failed because it removed the people whose judgment the AI needed to be trained on, then discovered too late that judgment cannot be conjured from a system prompt. The companies now growing fastest with AI are the ones who never made that trade in the first place.

The job market is not choosing between humans and AI. It is quietly punishing companies that forgot they needed both, and rewarding the professionals who make themselves the judgment layer AI still cannot replace.

Comments

comments