A year ago, "we're replacing that team with AI" was the line every executive wanted to say on an earnings call. Now some of those same executives are quietly posting the old jobs again — sometimes for the exact people they let go. The trend got a name in 2026: the AI boomerang.

On February 3, 2026, Gartner predicted that half of the companies that reduced customer-service headcount because of AI will rehire for similar roles by 2027, usually under different job titles. The reason is blunt: the AI wasn't ready to do the whole job. It handled the easy, repetitive tickets and then fell over on the messy, emotional, judgment-heavy ones, which turned out to be the part customers actually cared about.

What's actually happening

The cleanest example is Klarna, the buy-now-pay-later company. Between 2022 and 2024 it cut roughly 700 customer-service roles and handed the work to an AI assistant built with OpenAI, then froze hiring and told the market a chatbot was doing the work of those 700 people. In 2025, CEO Sebastian Siemiatkowski reversed course and started hiring humans again, telling Bloomberg the AI-only approach produced "lower quality" support and that the company had underestimated how much customers valued reaching a person.

IBM is the more interesting case, because the headline and the reality point in opposite directions. In early 2023 IBM signaled that AI would take over thousands of back-office roles, and CEO Arvind Krishna later said AI had replaced "a couple hundred" HR jobs through its internal AskHR tool, which automated about 94% of routine HR requests. But Krishna also said IBM's total employment went up, not down: the company redirected the money saved on HR into hiring more programmers and salespeople. The AI didn't shrink IBM. It moved people from one kind of work to another.

Then there's Duolingo, which is mostly a cautionary tale about talking about AI. In April 2025, CEO Luis von Ahn sent an "AI-first" memo saying the company would stop using contractors for work AI could handle and would grade employees on their AI use. The backlash was immediate (users threatened to delete the app), and von Ahn walked it back, admitting "I didn't do that well," then a year later dropped the AI-usage requirement from performance reviews entirely. Worth being precise here: Duolingo never laid off a full-time employee, and its headcount grew that year. What it cut was contractors, a real change, but not the mass AI layoff the memo implied.

Source spread

• Gartner — press release (Feb 3, 2026) [skeptic] — the primary forecast; based on an October 2025 survey of 321 customer-service leaders. Notably, only about 20% said they had cut agent headcount because of AI at all.
• Entrepreneur — Klarna reverses course [skeptic] — the CEO's own "lower quality" admission and the human-hiring restart.
• Entrepreneur — IBM's AskHR [builder] — the "couple hundred" HR roles and the total-headcount-went-up detail.
• Fortune — Duolingo drops the AI requirement [skeptic] — the full retreat from the AI-first stance.
• Inc. — Gartner's real lesson [builder] — analysis of why the reversals happen and what gets rehired.

What's real:

• The reversal is documented, not vibes. Gartner is a primary source, Klarna's CEO said it on the record, and IBM's own CEO described the redeployment. These are not anonymous "sources say" claims.
• AI reliably handled the routine tier and reliably struggled with the rest. AskHR automated 94% of requests; the remaining 6% — the sensitive, ambiguous, human ones — still needed people. That ratio is the whole story.
• Rehired workers have leverage. When a company has to undo a public AI bet, the people it needs back know it, and several reports note returning staff negotiating higher pay.

What deserves a side-eye:

• The "AI killed the jobs" framing was always shakier than the headlines. Gartner found only about 20% of support leaders actually cut headcount for AI; a lot of 2025's layoffs were ordinary cost-cutting wearing an AI costume.
• "Half will reverse by 2027" is a forecast, not a fact. It's a credible one from a credible firm, but it hasn't happened yet.
• Duolingo and IBM get lumped into "AI boomerang" lists where they don't fully belong — one cut contractors, the other grew. Precision matters here, because the sloppy version of this story is its own kind of hype.

Samwise's take

The satisfying version of this story is "AI flopped, humans win, executives humbled." The truer version is more useful: AI quietly ate the bottom 90% of a lot of jobs and choked on the top 10% — and that top 10% is where the judgment, the empathy, and the accountability live. Companies that fired the whole role learned they'd only automated part of it.

If you're a worker, the lesson isn't "you're safe." It's that the safe work is the hard-to-script work: the angry customer, the edge case, the call where the right answer is "let me actually look into this." If you're a builder or a manager, the boomerang is a warning label — the cheapest way to look foolish in 2027 is to fire a team in 2026 on the assumption a model can fully replace them, then pay a premium to hire them back when it can't.

What builders need to know

• Deploy AI as a tier-one filter, not a full replacement. The pattern that works is AI handles routine volume and escalates cleanly to a human; the pattern that boomerangs is AI as the only line of support.
• Measure the handoff, not just the deflection rate. Klarna's bot looked great on cost-per-ticket and bad on the thing that didn't show up in that metric: whether the customer felt heard.
• Keep the institutional knowledge in the building. The expensive part of the boomerang isn't the rehire — it's re-learning everything the laid-off people knew.

Further reading

• Gartner's forecast press release
• Inc.'s breakdown of the rehiring trend
• Klarna's reversal, in the CEO's words