Photo: Google Gemini · AI-generated
Corporate America spent the last two years running an experiment: replace headcount with AI, bank the savings, move on. The results are now coming in, and a growing number of household names are quietly reversing course.
The reversals are piling up
Ford leaned hard on automated quality-control systems to catch defects before vehicles reached the assembly line — and then spent the better part of three years hiring back 350 veteran “gray beard” engineers after the AI missed problems experienced inspectors would have caught immediately. The payoff: Ford topped JD Power’s 2026 Initial Quality Study for the first time since 2010 — after bringing humans back in, not after doubling down on automation.
Klarna told the market in 2024 that its AI assistant did the work of 700 customer-service agents, handling 75% of chats. It has since reversed course, rehiring human agents after satisfaction scores dropped and edge cases overwhelmed the bot. CEO Sebastian Siemiatkowski’s own words: “We went too far.”
Commonwealth Bank of Australia cut 45 customer-service roles after deploying a voice bot it said had reduced call volumes by 2,000 a week. A union challenge at the workplace tribunal revealed the opposite was true — volumes were rising, and staff were being pulled onto overtime to cover the gap. CBA reversed the redundancies and admitted an “error.”
IBM is running the same realisation in the opposite direction — tripling its US entry-level hiring for 2026 rather than cutting it. CHRO Nickle LaMoreaux put it plainly: the roles being added are “for all these jobs that we’re being told AI can do.” IBM is redesigning them around what AI can’t do — client-facing judgment, not routine execution.
The pattern shows up in the data too
This is not four anecdotes. Robert Half data shows 32% of US hiring managers who eliminated a role primarily because of AI have already rehired for that same or a similar position. Orgvue surveyed senior leaders across six countries and found 39% had made staff redundant specifically because of AI deployment — and 55% now admit it was the wrong call. Forrester’s research puts the regret rate at a similar 55%, and finds that for roughly a third of companies, the cost of fixing AI’s mistakes plus emergency rehiring wiped out the payroll savings entirely. Gartner now predicts that half of the companies that cut customer-service jobs citing AI will rehire for the same functions by 2027 — often under a different job title, which is its own quiet admission.
Where the line actually sits
None of this means AI automation failed. Every one of these companies still uses AI extensively, and in the routine, well-scoped parts of the work, it performs well. What it consistently cannot do — not yet, and arguably not by simply scaling up — is exercise judgment in situations nobody wrote a rule for: a defect pattern an experienced inspector would flag on sight, a customer conversation that needs empathy instead of a script, an HR process that depends on institutional memory. The companies above did not discover that AI is bad. They discovered that removing the human who catches the exception is far more expensive than the salary it was meant to save.
Why this matches what we already built
We wrote about the cost side of this same mistake a few weeks ago — Gartner’s forecast that AI coding costs will overtake developer salaries by 2028, driven by the same instinct to remove human oversight in the name of efficiency. The current rehiring wave is the workforce version of the same story.
At exbisoft, we never treated AI as a replacement for engineering judgment — we treat it as a tool our engineers use, not one that uses them. That is also why our application support and maintenance work is staffed by people who stay on the system long after launch, not a support queue that hands edge cases back to a model. The companies above are learning this the expensive way. It is cheaper to build it right from the start.
Rethinking where automation actually belongs in your operation? Talk to us.