Why Companies That Cut Engineers Are Quietly Hiring Them Back

Ford's automated quality-control software was supposed to do what an engineer used to do: catch a design flaw before it reached someone's driveway. It didn't catch enough of them. Ford has reportedly rehired, newly hired, or promoted 350 experienced engineers to close that gap by hand, and its vehicles topped JD Power's 2026 Initial Quality Study for the first time since 2010.Ford's rehiring, newly hiring, or promoting 350 experienced engineers helped it top JD Power's 2026 Initial Quality Study rankings for the first time since 2010. That single reversal is a decent stand-in for what's happening across a lot of engineering orgs that cut deep in 2023 and 2024.

More than 260,000 tech workers lost their jobs in 2023 alone, with many companies explicitly citing AI automation as the justification. The pace hasn't slowed. One live tracker counts 383 layoff events touching 210,741 workers so far in 2026, and announced tech layoffs are running roughly 83 percent higher year over year through June. Amazon has cut up to 30,000 corporate jobs, attributing the reductions in part to AI-driven efficiencies across cloud, HR, and operations. Meta cut around 8,000 more employees in May, on top of more than 21,000 already gone in earlier rounds. Microsoft eliminated 4,800 positions this year and, notably, said the quiet part out loud in the opposite direction: its chief people officer stated the roles eliminated were not being replaced by AI. That's the honest version most companies skipped when "AI" made the more convenient headline.

Forrester's 2026 Future of Work report: 55 percent of employers now regret their AI-driven layoffs. Gartner projects that by 2027, half of companies that cut roles for AI will rehire for similar functions.

The rehiring numbers back this up from a different angle. Robert Half found that 29 percent of organizations that cut employees due to an AI-related reduction had already rehired into the positions they cut. A separate Orgvue study, looking only at leaders who'd made cuts specifically for AI, found that 55 percent of them now admit the decision was wrong. Different firms, different survey designs, and the number keeps landing in roughly the same place.

This isn't a story about AI failing to do anything. It's a story about what got counted, and what didn't.

IBM's HR reorganization is the cleanest version of this, mostly because the company let the actual numbers become public. Its AI system handled around 94 percent of routine HR requests but was unable to meet the other 6 percent, which included ethical dilemmas. That 6 percent doesn't sound like much until you notice the people who used to handle it were the ones who'd been let go. IBM's chief human resources officer, Nickle LaMoreaux, later warned that skipping entry-level hiring for too long means the pipeline simply dries up, and the company announced plans to triple its U.S. entry-level hiring across all business units in 2026.

Here's the mechanism most coverage skips past: companies measured the wrong percentage. The question they asked was "what share of this role's tasks can AI complete?" and the answer, 94 percent, 96 percent, whatever the demo showed, was genuinely impressive. The question that actually predicts whether a cut is safe is different: what share of this role's consequential failures does a human catch before they reach a customer, a regulator, or a production system? Those two numbers look similar in a pilot. They diverge violently at scale, because the failures that matter are, almost by definition, the ones nobody wrote a training example for yet.

The Rehiring Tax: the premium a company pays, in salary, ramp time, and lost institutional judgment, to buy back the exact engineering capacity it gave away for free.

It shows up in three places. Returning workers are commanding salary premiums of 20 to 35 percent above pre-layoff levels, because the market has repriced skills these organizations proved they can't function without. One-third of employers who reversed the layoffs spent more on restaffing than they originally saved, once recruiting, onboarding, and salary bumps are counted. And the part no invoice captures: the specific context a departed engineer carried (which client's system breaks in which way, which 2022 decision not to repeat) doesn't rehire with them. A new person, however senior, starts that part from zero.

The tax isn't a penalty for using AI. It's a penalty for mistaking a completion rate for a coverage guarantee.

This isn't the first gap we've written about between what AI visibly does and what it invisibly skips.

Ninety-six percent of developers don't trust AI-generated code without manual review, according to SonarSource's 2026 State of Code Report. That statistic makes no sense if AI had actually replaced the reviewer. It makes complete sense if AI replaced the typing, and companies mistook that for replacing the reviewing, then discovered the reviewer was the position they'd eliminated.

None of the four companies above skipped AI. They skipped the step where someone asks which slice of the job doesn't have a template yet, and who's going to own it once the person who used to is gone.

The fix isn't "stop using AI," and it isn't "never reduce headcount" either. Plenty of the layoffs in the numbers above were legitimate corrections to pandemic-era overhiring that had nothing to do with AI at all. The fix is asking a more specific question before the cut, not after the rehire.

Three questions predict whether a role is actually safe to remove:

1. 1.What's the failure rate at production volume, not in the pilot? Ten examples in a demo and ten thousand in production are different tests. Ask for the second number before the first one gets used to justify anything.
2. 2.Who currently catches the edge cases, and where does that judgment go if they leave? If the honest answer is "nowhere, it just stops happening," that's the slice of the job nobody's priced yet.
3. 3.Is today's savings smaller than the cost of reversing the decision in twelve months? Twenty to thirty-five percent above prior salary, plus recruiting, plus ramp time, is the real comparison, not the payroll line being cut.

Some organizations are already running this calculation correctly: piloting smaller, senior-heavy teams instead of wholesale cuts, and holding budget for AI-specific roles rather than assuming the remaining staff absorbs the gap for free. The difference between them and Ford eighteen months ago was never the technology. It was whether anyone asked what the last slice of the job actually costs before deciding it was the cheap part to cut.

The companies quietly rehiring right now didn't discover that AI doesn't work. They discovered what the last six percent of any job actually costs, and that the only place to price it accurately was before they let the person go, not after.

If you're staring at a headcount decision this quarter, get a second read on which slice of the job you'd be cutting before the market reprices it for you.