In July 2025, the Commonwealth Bank of Australia declared 45 roles in its Customer Service Direct business redundant, and pointed at a voice bot. "Our investment in technology, including AI, is making it easier and faster for customers to get help, especially in our call centres," the bank said at the time, as the ABC reported. Its staff told the finance workers' union the opposite was happening. Call volumes were rising, the union said, "with management scrambling to offer overtime and even pulling team leaders onto the phones."
That case has become the cheerful headline of the summer: companies cut people for AI, the automation underdelivered, and now they are quietly hiring the same people back. The staffing firm Kelly Services put three names to it in an August 3 briefing, stating that "major companies including Ford, Commonwealth Bank, and IBM are reversing AI-driven layoffs after automation failed to deliver expected results," and pairing that with a survey figure: "32% of U.S. hiring managers eliminated a role due to AI and later rehired for the same or similar position." I went through all three companies at source, and through the survey, working only from published documents, filings, and the companies' own words. I asked none of them to comment and I interviewed nobody. One of the three is a genuine reversal. The other two are not, and I found no methodology published anywhere for the survey figure.
The bank really did back down, a month after announcing the cuts
Commonwealth Bank holds up. The Finance Sector Union, which represents Australian finance workers, took the redundancies to the Fair Work Commission, the country's workplace tribunal, challenging the bank's account of call volumes. Its members were counting. On August 21, 2025, the bank reversed. CBA said its assessment "did not adequately consider all relevant business considerations" and that "we should have been more thorough in our assessment of the roles required." Staff were given the choice to stay in their roles, be redeployed, or leave. How many took which, the bank has not said.
No ruling by the Commission appears in anything I could find. CBA reversed while the dispute was on foot and gave its own reason, which was that it had assessed the roles badly. Bad software gets tolerated for years. What was unusual here was a union with a tribunal date and members who could contradict the company's own numbers from the phones. The union's national secretary, Julia Angrisano, put the bank's motive less generously. "CBA has been caught out trying to dress up job cuts as innovation," she said. "Using AI as a cover for slashing secure jobs is a cynical cost-cutting exercise, and workers know it." The union's own account of the reversal, headlined "WIN: CBA backflips on customer service job cuts, admits they got it wrong," is open only to members past its opening lines; the Angrisano quotes above are the ABC's.
Now set those 45 roles against the year they were cut in. CBA reported a record cash profit of 10.25 billion Australian dollars for the 2025 financial year, the ABC noted in the same report, which also carried the chief executive, Matt Comyn, saying about 2,000 extra staff had been hired in recent months and acknowledging that many of those roles were added in India.
And look at what followed the good news. By this May the bank's chatbot was resolving almost nine in every 10 customer conversations without a human, Information Age reported in July. In the same piece, citing Bloomberg reporting I could not read directly, it said hundreds of chat support roles had been cut at a Johannesburg outsourcing firm that staffed CBA's online chat. The union says the bank also moved to eliminate a further 176 technology and engineering roles, alleging that some of those were then advertised through CBA's India-based subsidiary, and that around 800 CBA roles went in a single year. Whether the original 45 sit inside that 800 or beside it, the union does not say.
So 45 roles came back. That is the round trip at its actual scale, and it is worth being precise about where the work went instead: to a chatbot, to Johannesburg, and to India. The jobs did not evaporate into automation. They moved.
Nothing in what I found says Ford's engineers or IBM's HR staff were cut for AI
Ford's version reads like a reversal, and one of its own executives used reversal language. Chief operating officer Kumar Galhotra told reporters the company had been "relying more and more on automated quality systems" and so "brought back technical specialists," in remarks TechCrunch relayed from Bloomberg's report of the call. Charles Poon, Ford's vice president of vehicle hardware engineering, was blunter on the same call, in remarks Fox Business carried from that same Bloomberg report: "Mistakenly, we thought that by just introducing artificial intelligence and ingesting the design requirements that we had, that would produce a high quality product."
Brought back is the strongest fact against the reading I am about to give, so weigh it properly. Ford's written accounts never use it. Ford's own post says it "hired roughly 300 veteran engineers" who now act as internal auditors running weekly design reviews. A Ford spokesperson told Fortune the company "hired more than 350 experienced tech specialists." Hired, twice, in writing, on the record. And neither the call nor the written accounts says a single one of these people was cut because of AI. Fox Business, summarizing the rest of what Poon said on that call, reports that many of Ford's veteran technicians left the company before their knowledge could be used to train the tools. Left. Fortune says the 350 are "both former Ford employees and workers from suppliers." Hiring back people who retired or went to a supplier is a rehire in the ordinary sense of the word. It is not the reversal of an AI-driven layoff, and no source I found other than the briefing under test says Ford made one.
Ford also does not accept the premise that AI failed. Its own post credits AI vision systems on the plant floor and hundreds of thousands of automated test scenarios as part of the fix, and says the effort made it the top mainstream brand in this year's J.D. Power quality study, second overall among corporations and third among brands. Meanwhile Detroit's three automakers have cut more than 20,000 US salaried jobs from their recent peaks, which CNBC puts at 19 percent of their combined workforces, working from the companies' own filings. Ford alone was down roughly 5,300 from its 2020 salaried peak last year, to about 30,700. Its chief executive, Jim Farley, told the Aspen Ideas Festival in the summer of 2025 that AI "is going to replace literally half of all white-collar workers in the U.S.," in remarks CNBC reported in the same piece. Ford's salaried headcount is down roughly 5,300 from its 2020 peak to 2025, against 300 veteran engineers hired back over what Ford calls the last few years. That is not a company reversing course. It is a company buying back one specific thing it should not have let walk out the door.
IBM's case is thinner still, and the join is a single word. Kelly's briefing says IBM replaced HR functions with AI that handled 94 percent of routine requests but "stumbled on the remaining 6%" of them, ethical dilemmas included, and "subsequently announced plans to triple U.S. entry-level hiring in 2026." Both halves appear a month earlier, in CNBC's July 1 piece, which sets them in consecutive sentences and cites nothing for either: the AI "was unable to meet the other 6%, which included ethical dilemmas," and "IBM then announced plans to triple its U.S. entry-level hiring across all business units in 2026." Kelly's briefing lists CNBC among its sources. Behind CNBC I could find nothing: no IBM statement describing a stumble, and no account anywhere linking the automation to the hiring plan. AskHR, IBM's internal HR assistant, does handle about 94 percent of employee questions, and its rollout went badly before it went well: HR's internal satisfaction score fell from clearly positive to badly negative when the team switched off the HR email address and phone number overnight (from +19 to -35), and now sits well above where it started (+74). The hiring plan, announced in February, is a pipeline argument rather than a retraction, and a good one: "The entry level jobs that you had two to three years ago, AI can do most of them," said chief human resources officer Nickle LaMoreaux. "You have to rewrite every job." Entry-level, at IBM, means recent graduates plus career changers and people returning to work, a spokesperson told Axios. And in 2025 IBM cut about 1 percent of its 270,000 staff, which Axios reports was driven by "business demand," with no mention of AI.
One source does say IBM rehired, and it is one I lean on elsewhere here. In a single line of its August 21, 2025 report on the CBA reversal, Information Age wrote that IBM "rehired staff after replacing much of its human resources division with AI." I could find nothing behind that sentence: no IBM statement, no named role, no number. It is Kelly's claim again, asserted the same way, in a publication I otherwise trust.
Watch the clause fall off the 32 percent
Which leaves the number. It comes from Robert Half, a firm that places contract and permanent staff for a fee, and that is worth holding in mind, because a finding that companies are rehiring is a finding that companies need Robert Half. I found no US release from the firm carrying the figure, and no methodology for it anywhere. It sent the data to CNBC and provided it to Fast Company, and published no sentence of its own that I could find until August 11, in a UK trade release. By then three organizations had already put the number in front of readers, in three different forms. Here they are in the order they appeared.
Fast Company, June 5: "According to new research from the consulting firm Robert Half that was reviewed by Fast Company, nearly a third (32%) of hiring managers say that their organizations eliminated a role or let someone go primarily due to productivity gains from AI or automation, only later to rehire for that exact role."
CNBC, July 1: "Meanwhile, 32% of U.S. hiring managers said they eliminated a role primarily due to AI and later rehired for the same or a similar position, according to data from Robert Half sent to CNBC."
Kelly Services, August 3, in its summary: "Major companies including Ford, Commonwealth Bank, and IBM are reversing AI-driven layoffs after automation failed to deliver expected results; 32% of U.S. hiring managers eliminated a role due to AI and later rehired for the same or similar position." The briefing runs the figure again further down as "primarily due to AI," without the clause both times.
Robert Half, August 11: "Robert Half research found 32% of U.S. hiring managers and 34% of Canadian hiring managers who eliminated a role primarily because of AI later rehired for the same or a similar position."
The firm's own sentence, the one that arrives last, is the only one carrying the clause "who eliminated a role primarily because of AI," and that clause restricts the base to managers who had already made such a cut. Without it the sentence reads as a third of all hiring managers, which is what the first three versions say. Fast Company's is also the only version that says "that exact role" rather than "the same or a similar position," a tighter claim on thinner ground. Nobody lied. A restrictive clause is simply missing from the three versions that reached readers first, and it was the clause carrying the whole meaning. Which of them matches what Robert Half supplied, only Robert Half can say.
I found no Robert Half document publishing the sample size behind that question, the field dates, the wording put to respondents, or a margin of error. The only methodology anyone has printed is the sample size Fast Company was given: 2,000 US hiring managers, with a breakdown by sector. A separate July release from the firm describes a survey of more than 2,000 US hiring managers conducted in April by an unnamed independent research firm, but does not carry the rehiring figure. I checked the company's newsroom index on August 20 and found no release that does.
The outside numbers are more modest and more useful, though apply the same test to them and neither firm is disinterested: Orgvue sells organizational design software, Gartner sells advisory. The much-cited finding that 55 percent of employers regret their AI redundancies is from Orgvue in April 2025, not this year, and it is 55 percent of the 39 percent who made such cuts, so about a fifth of the sample. Fast Company attributes the same 55 percent to Forrester Research instead, which is its own demonstration of the problem. And Gartner's own field data, from 321 customer service leaders surveyed last October, found that only 20 percent had actually reduced staffing because of AI. Its analyst Kathy Ross put the rest plainly: "Most recent workforce reductions were influenced by broader economic conditions rather than automation alone."
Ask who is selling the comeback
While the reversal story was being written up, Visa cut about 2,600 people, roughly 7 percent of its workforce, and its chief executive told staff that AI "is also helping to accelerate this evolution and shape the way work gets done at Visa." CNBC, which confirmed the memo, also carried the caveat: a person with direct knowledge of the matter said AI played a significant role in the layoffs but "wasn't the sole driver."
Visa's quarterly filing, for the three months ended June 30 and so for a period that closed a month before the cuts were announced, books 563 million dollars of severance as a special item and gives its own reason: "actions taken to drive operational efficiencies and reinvest in high-growth opportunities." The filing does not name AI, and it does not itself connect that charge to the July announcement. HR Executive is the one that draws the line, describing the 563 million as severance "tied to the workforce reduction," and attributing that description to Visa's own earnings materials, which carry the figure but do not make that connection. Revenue was up 14 percent in the quarter. Profit was up 7 percent. Three audiences, three vocabularies, and none of them a company discovering that AI cannot do the job.
So notice who is telling you the round trip is happening. Kelly Services, which sells staffing, publishes the briefing naming the three companies. Robert Half, which sells staffing, supplies the number, has not published how it got it, and already has a name ready for what it expects next. It calls it correction recruitment. Turn that phrase over and look at the underside. A layoff becomes a correction, an error of calibration rather than a decision somebody signed, and the worker becomes the correction, to be recruited back at whatever the market now says they are worth, by the firm that bills for placing them.
So when the next briefing tells you the machines failed and they need us after all, ask it the question it never asks: not whether the automation worked, but who made anyone give the job back. On this record there is exactly one case where somebody did, and the mechanism is on file. Call center workers counted the calls, their union carried the count to a tribunal, the bank wrote down that it had been wrong, and 45 roles came back. No briefing and no survey will do that part for you.
