AI-cited layoffs just hit a record, and the reversals are catching up to them
US employers announced 97,006 job cuts in May 2026, the highest May total since 2020 and up 16% from April, according to Challenger, Gray & Christmas. AI was cited as the reason for 38,579 of those cuts, or 40% of the total, up sharply from 7% in January. It was the highest monthly total for AI-cited cuts since Challenger began tracking the reason in 2023, and the third straight month AI led every other reason employers gave. Year to date, AI has been cited in 87,714 cuts, already more than all of 2025 combined. At the same time, CNBC reported on July 1, 2026 that several of the employers behind these cuts are already walking parts of them back. Ford is reportedly rehiring hundreds of experienced engineers after its automated quality systems fell short. Commonwealth Bank of Australia reversed a decision to replace customer service staff with an AI voice bot after call volumes rose instead of fell. And IBM, after replacing much of its routine HR casework with AI, announced plans to triple its US entry-level hiring in 2026. This is not an isolated anecdote. CNBC cited data from Robert Half showing that 32% of US hiring managers who eliminated a role primarily because of AI have already rehired for the same or a similar position. Employers are cutting AI-attributed roles at a record pace, and a meaningful share of those same employers are re-hiring within months.
Why IBM's version of this story is the one worth studying
Most AI layoff stories stop at "the automation didn't work." IBM's story is more specific, and more useful, because it names the exact line where automation stopped working. IBM's HR AI handled about 94% of routine requests successfully. The remaining 6% involved ethical dilemmas and judgment calls the AI could not reliably make. A 94% success rate is a real result. It is not a failure story. The mistake IBM avoided was concluding that a 94% automation rate meant the business needed fewer people going forward. IBM's chief human resources officer, Nickle LaMoreaux, framed the decision at a Charter AI Summit in New York this way: "If we don't continue to invest in entry-level hires, what happens in 3 to 5 years? There's no pipeline; the well simply dries up." The entry-level roles IBM is now expanding are not just where routine work happens. They are also where the people who will eventually handle that hard 6% get their training. Cut the entry-level role because AI now does most of its routine content, and you have not just cut a cost. You have cut the training ground for the judgment your business will need in a few years, right when the easy 94% is fully automated and only the hard part is left.
What this is not: a case against automating, or evidence AI does not work
It would be easy to read this the wrong way, as proof that AI cannot be trusted with real work. That is not the finding here. IBM's AI is still doing the 94% of HR requests it was built to handle, and that work is not coming back to a person. The finding is narrower and more useful: know which specific tasks in a role are routine and safe to automate, which require judgment, and who builds that judgment if the entry point to the role disappears. IBM is not acting alone on this. Teneo's Vision 2026 CEO and Investor Outlook Survey, published in December 2025 and based on responses from more than 350 global CEOs and 400 institutional investors, found that 67% of CEOs expect AI to increase entry-level headcount in 2026, not reduce it, and 58% expect an increase in senior leadership hiring as well. That does not mean layoffs are slowing. Challenger's numbers show the opposite. It means a meaningful share of large employers are treating "which roles do we cut" and "who trains our future judgment" as two separate questions, not one. The mistake worth avoiding is not automating. It is treating a role as a single line item to eliminate once AI can do most of it, without asking what specific piece of that role was actually producing your business's future judgment.
The operational lesson: map the task, not the role
IBM Think profiled a practical version of this idea from Matt Beane, a UC Santa Barbara professor who built a framework called SkillBench for exactly this problem. Instead of deciding whether to automate a role as a whole, Beane maps a team's actual work into small task packets and scores each one on two axes: how much productivity gain AI offers on that task, and how much skill-building value the task has for the person doing it. Tasks that score high on AI productivity and low on skill-building are what Beane calls "skill deserts": safe to automate fully, because no one was learning much from doing them anyway. Tasks that score high on both are the ones worth protecting, because automating them removes the exact experience that turns a junior person into someone who can handle judgment calls later. A small or medium business does not need Beane's full framework to use the idea. Before eliminating a junior role because AI now covers most of its output, list what that person actually does today. Separate the routine, repeatable pieces from the pieces where they are learning to make a judgment call, handle an exception, or work directly with a client or a difficult case. Automate the first group. For the second group, the better move is usually to redesign the role around it, not eliminate the role, similar to how IBM describes its entry-level developers now spending less time on routine coding and more time working directly with customers.
What a serious business should do next
If your business has already cut, or is considering cutting, a role because AI now handles most of its routine output, take an hour to list what that role actually did before you finalize the decision. Separate the tasks AI already does reliably from the tasks that involved judgment, exceptions, or direct client contact. If the second group is thin or nonexistent, the automation case is straightforward. If it is not, the real question is not whether to automate the routine part. It is who develops the judgment for the harder part once the role that used to teach it is gone. If you have already made a cut like this and are starting to see the signs Ford and Commonwealth Bank of Australia saw (issues slipping through, no one available who understands the edge cases, more escalations than expected) that is the same signal IBM responded to before it became a bigger problem. It is worth a direct review of what that role's harder 6% actually required, and whether anyone in the business can currently do it. For businesses without an obvious answer to "who trains the next person who can handle the hard cases," the fix is rarely to freeze all automation. It is usually a smaller, redesigned junior role built around oversight of AI output, exception handling, and direct contact with the hardest cases, the same shift IBM, McKinsey, and Cognizant have each described making to their own entry-level hiring in 2026.
The Atlacis view
The record-setting Challenger numbers and the reversal stories are two sides of the same decision problem. Businesses are cutting roles based on what AI can now do, and a meaningful share of them are finding out afterward that the role was doing more than its routine output, whether that is Ford's design judgment, Commonwealth Bank's call handling under real volume, or IBM's hard 6% of HR cases. The fix is not to slow down on AI. IBM's AI is still handling 94% of the work it was built for, and that is a real result worth having. The fix is to make the automation decision at the task level instead of the role level, so the parts of a job that build judgment do not get eliminated along with the parts that were always routine. At Atlacis, we help business owners map that distinction before a role gets cut, not after. If you are weighing whether AI has genuinely covered a role or just its routine 90%, the right first step is a clear look at what that role actually does today and what happens to the harder cases once the automation is in place.
The short version
- AI was cited as the reason for 40% of US layoffs announced in May 2026, the highest monthly share since Challenger, Gray & Christmas began tracking the reason in 2023, and the third straight month AI led all cited layoff reasons.
- CNBC reported on July 1, 2026 that Ford, Commonwealth Bank of Australia, and IBM are all rebuilding human capacity in work they had shifted to AI. Robert Half data shows 32% of US hiring managers who cut a role for AI have already rehired for the same or a similar position.
- IBM's AI automated about 94% of routine HR requests but could not reliably handle the remaining 6%, cases involving ethical judgment and nuanced decisions.
- IBM responded by announcing plans to triple US entry-level hiring in 2026, arguing that entry-level roles are the pipeline that produces the judgment a business needs in 3 to 5 years, not just a source of routine output.
- A Teneo Vision 2026 survey of more than 350 global CEOs found 67% expect AI to increase entry-level headcount in 2026, not reduce it, suggesting IBM's approach reflects a broader planning shift among large employers even as overall AI-cited layoffs rise.
- The practical framework: map a role's tasks by AI productivity gain and by skill-building value. Automate the routine 'skill desert' tasks. Redesign, rather than eliminate, the tasks that build the judgment your business will need later.
Where ATLACIS can help
Sources
- CNBC (Justina Lee): Employers who laid off workers citing AI are already starting to regret it (July 1, 2026)
- Challenger, Gray & Christmas: May 2026 Job Cut Report, AI drives May cuts to 97,006 (released June 4, 2026)
- IBM Think: The bottom rung returns as AI reshapes entry-level jobs
- Teneo: CEO and Investor Confidence Remains Strong Heading Into 2026 Despite Global Headwinds (Vision 2026 survey, December 15, 2025)