AI completes tasks, not jobs: why the rehiring wave was predictable
Companies are rehiring the people they laid off for AI. That was never going to be a surprise.
The news nobody prepared for
Ford brought back hundreds of experienced engineers this year to fix quality problems automation could not solve. IBM automated 94% of AskHR requests with AI. IBM is now tripling entry-level hiring in 2026, because the remaining 6% needed a person. Klarna walked back its customer service cuts after the AI chatbot could not carry the full load. None of these are isolated stories. They are the same pattern, repeating at three different companies.
The market noticed. One in three US hiring managers eliminated a role citing AI. Most later rehired the same role or a comparable one, according to Robert Half. 55% of leaders who cut jobs in an AI rollout now say it was a mistake. That is per Orgvue, the workforce-planning research firm. Gartner projects that by 2027, half of the companies that cut customer service roles for AI will rehire similar functions.
The wrong diagnosis
The easy explanation is that AI overpromised and companies got burned. The data says something else. Gartner found that 80% of large enterprises reduced headcount after AI projects. There was no correlation between the cuts and AI return on investment. High-ROI companies and low-ROI companies cut jobs at the same rate.
That number kills the "AI failed" explanation. If AI performance did not predict who got cut, AI performance was never the basis for the decision. The cuts were a budget line reacting to a headline, not an engineering decision reacting to a system's actual capability. Blaming the technology hides the real failure. Companies spent on AI without defining what the system was supposed to do. They never decided which parts of the job stayed human.
The 6% that explains everything
IBM's AskHR case is the clearest data point available. Automating 94% of a workflow is a real result. It is also proof that the remaining 6% was never going to close with a bigger model or a longer prompt. That 6% is judgment. It is the case that does not match the pattern. It is the exception that needs context, the decision with an ethical weight attached to it.
Every AI system we have shipped into production has a version of this line. The ratio changes by task. The existence of the line does not. A company that designs for 100% automation from day one is not scoping a system. It is scoping a layoff, and the system will eventually prove it wrong.
Our engineering point of view
AI completes tasks. It does not complete jobs. A job is a bundle of tasks plus judgment, context, and accountability for the outcome. AI takes the tasks. The rest stays with a person, and that is not a limitation to design around later. It is the architecture.
Human-in-the-loop is not a compliance checkbox added after a pilot goes sideways. It is a decision made on day one, alongside the model choice and the data pipeline. The decision covers three things: which choices the system makes alone, which ones route to a person, and who that person is. We build the escalation path before we build the automation. A system without a defined escalation path has no owner when it hits its 6%. It has an incident instead.
The companies now rehiring skipped that step. They automated the workflow and cut the headcount the workflow used to require. Then they found out in production which decisions actually needed a person. The companies that designed human-in-the-loop into the system from the start are not rehiring. They never cut the roles the system was never going to replace.
The right question
The question driving the last two years of AI headcount decisions was "how many people can this replace." That question produced the layoffs, and the layoffs produced the rehiring, and the rehiring is now producing the headlines. It was the wrong question from the start.
The right one is what a team can achieve with the system doing the tasks it is good at. A person handles the 6% that requires judgment. That question does not make for as clean a press release. It is the one that keeps a team the same size in year three. It beats cutting the team in year one and rebuilding it in year two.
Getting that question right on day one is a readiness problem, not a technology problem. That is where we turn next.
Details
August 26, 2026
Applaudo
Name: José Giammattei, Co Founder and CEO
Phone: +503 7797-6586
Email: jgiammattei@applaudo.com