Deciding in a Market That Keeps Reversing Itself

Career Landscape Reports: 4 of 5 in a series

In the same week this year, it was possible to read a headline about a company eliminating hundreds of jobs and citing AI as the reason, and a separate headline about a different company quietly rehiring the people it had let go for the exact same reason twelve months earlier. Both headlines were accurate. Both were describing the same underlying market. That’s the actual condition anyone making a career decision right now has to plan around, and it’s a harder problem than either the doom coverage or the reassurance coverage tends to admit: not “is AI taking jobs,” but “the signal keeps changing direction, and I have to decide anyway.”

The first three pieces in this series gave you tools for reading that signal: where AI genuinely expands work, where the traditional path into that work is narrowing, and how to size up your own specific exposure. None of those tools tell you what to actually do when the market itself won’t hold still long enough for a clear answer to matter. This piece is about that problem directly.

The Whiplash Is Real, Documented, and Ongoing

Start with how common reversal actually is, because the scale of it is easy to underestimate from scattered headlines.

Gartner’s Customer Service and Support practice, in a forecast released in February 2026, predicted that by 2027, half of the companies that attributed headcount reductions to AI will rehire staff to perform similar work, typically under different job titles. Gartner’s own analyst on the report, Kathy Ross, put the underlying cause plainly: many of the original cuts were driven more by general economic pressure than by AI actually being ready to do the work, and companies are now running into the limits of automation on exactly the tasks that require judgment, empathy, and handling situations nobody scripted in advance.

Forrester’s Predictions 2026 report found that 55% of employers who made AI-driven cuts now regret the decision. A February 2026 survey of 600 HR leaders by the outplacement firm Careerminds found the regret was already turning into action: two-thirds of companies that had made AI-driven cuts were already rehiring, more than a third had brought back over half the roles they’d eliminated, and 52% of the rehiring happened within six months of the original cut. Nearly a third of those companies ended up spending more on rehiring than they’d saved by automating in the first place, meaning a real share of last year’s layoffs left the company worse off than if they’d never made the cut at all.

The individual company stories are specific and public. IBM deployed an AI system to handle HR requests that successfully resolved roughly 94% of incoming issues, and then discovered the remaining 6%, disproportionately situations requiring ethical judgment, exposed the actual limits of full automation clearly enough that the company announced plans to triple its U.S. entry-level hiring across every business unit in 2026. Ford and Commonwealth Bank of Australia both reversed AI-driven job cuts after concluding the technology couldn’t fully replace the roles they’d eliminated; Commonwealth Bank’s own statement acknowledged it “did not adequately consider all relevant business considerations” before the original cuts. Klarna, whose 2023 announcement that an AI assistant was doing the work of roughly 700 customer service agents became one of the most cited examples of AI displacement, brought human staff back and reached breakeven for the first time in May 2026, with its CEO telling the market directly that customers needed to know a human would always be reachable.

None of this means the layoffs were fake or the AI didn’t do anything. Gartner’s own May 2026 survey of 350 executives at large companies, all actively piloting or deploying AI, found that 80% had reduced headcount, and found no correlation between those reductions and any measurable return on the AI investment behind them. Companies with strong AI returns cut staff at nearly the same rate as companies with weak or negative returns. AI-attributed layoffs hit their highest single-month total on record in May 2026, at nearly 40% of all announced job cuts that month. The cutting and the rehiring are not two different phases of one story. They’re happening at the same time, often inside the same industries, sometimes inside the same companies a year apart.

The Understandable Response, and Why It Has a Cost Too

Faced with a signal that keeps reversing, a lot of people have landed on the same strategy: don’t move. Stay exactly where you are and wait for the picture to clarify.

There’s a name for this now. Job hugging describes employees staying in their current roles specifically out of anxiety rather than satisfaction or loyalty, and it’s grown fast. A February 2026 survey found 57% of U.S. workers identified as job huggers, up twelve points from just five months earlier. Among people who described themselves this way, 70% said they were specifically worried AI would affect their job within the next six months, and 63% feared layoffs in that same window. Separate research from MetLife found that 56% of employees were staying with their current employer out of necessity rather than genuine engagement, against a backdrop of employee financial confidence at its lowest measured level since 2012.

The instinct is entirely reasonable given everything in the previous section. What’s worth being honest about is that staying still isn’t actually a neutral, risk-free choice, even though it feels that way. The same research found that over half of self-identified job huggers reported working longer hours than before, and meaningful shares reported missed promotions and raises they’d otherwise have been positioned for. Freezing in place is itself a decision, with its own real cost, made under exactly the same uncertainty as any other option. It just doesn’t feel like a decision because it doesn’t feel like doing anything.

A Better Question Than “What Will Happen”

The mistake underneath both overreacting to a scary headline and freezing indefinitely is the same one: treating the goal as correctly predicting what the market will do next, when the market itself is demonstrably bad at predicting that about itself. Companies with access to their own internal numbers, deploying their own AI systems, are getting this wrong roughly half the time by their own later admission. There’s no reason to expect an outside individual, reading the same public headlines everyone else is reading, to do meaningfully better at forecasting.

A more useful question, borrowed from a decision-making framework popularized at Amazon and now applied fairly widely to individual career choices, is not “will this turn out to be the right call” but “how hard would this be to walk back if it isn’t.” The framework sorts decisions into two kinds. A two-way door is a decision you can reverse or adjust without major cost if it doesn’t work out: trying a new skill, taking on a stretch assignment, having an exploratory conversation with a company in a different field, updating how you present your experience. A one-way door is a decision that’s genuinely difficult or expensive to undo: quitting a stable role with no plan in response to a single alarming headline, or making an irreversible bet on one narrow tool or trend as your entire professional identity.

The practical value of this distinction, specifically in a market prone to reversing itself, is that it changes how much weight any single data point deserves. A two-way door doesn’t require you to be right about where the whole field is heading before you act on it, because you can correct course cheaply if the read turns out to be wrong. Reserve the deep deliberation, the waiting for more evidence, the reluctance to move without confidence, for the choices that are actually hard to undo. Most of the moves available to someone worried about their exposure right now, exploring adjacent skills, testing interest in a different specialization, having conversations before you need to, building visible evidence of judgment-level work, are two-way doors even though they don’t feel like it in the moment.

Treating a Single Announcement as Weak Evidence, on Purpose

There’s a second, related habit worth adopting deliberately: given how often this year’s confident AI announcement became next year’s quiet reversal, a single company’s public narrative, including your own employer’s, is weak evidence about where a field is actually heading. That’s not cynicism. It’s just matching your confidence to what the data has actually shown about how reliable these announcements have been.

In practice, that means waiting for a second data point before treating a splashy claim as settled fact, in either direction. If a company announces AI has eliminated an entire function, the honest response is curiosity about whether that holds twelve months later, not immediate alarm about what it means for your own adjacent role. If a company announces it’s tripling entry-level hiring in response to AI’s limits, as IBM did, the honest response is the same measured curiosity, not immediate relief. Individual announcements, in a market this prone to reversal, are data points in a longer pattern, not conclusions on their own.

Where This Leaves the Series

The tools in the first three pieces of this series are still the right ones for reading where you actually stand. This piece doesn’t replace them. It answers the question they don’t: given an honest read of your exposure, and given that the market supplying new information about that exposure keeps contradicting itself, you don’t need to wait for the contradictions to resolve before you act. You need a way to act that doesn’t depend on the market holding still, which mostly means favoring moves you can reverse over moves you can’t, and treating any single headline, including the reassuring ones, as one data point rather than the whole story. The final piece in this series turns from how to navigate this volatility to where the actual value is moving as it settles, and what building toward it deliberately looks like in practice.

I use AI tools to help research sources and draft these posts. I review, edit, and take responsibility for everything published here, including any errors.

---

Sources

This piece drew on the following research and reporting (several entries are secondary reporting on the underlying studies, not the original reports themselves):

• Gartner, “Gartner Predicts Half of Companies That Cut Customer Service Staff Due to AI Will Rehire by 2027,” press release, February 3, 2026, featuring commentary from Senior Director Analyst Kathy Ross.
• The Agency Recruiting, “AI Layoffs and Job Cuts: What They Mean for the Workforce,” 2026. Reporting on Gartner’s May 2026 survey of 350 executives finding no correlation between AI-driven headcount reductions and ROI.
• International Business Times UK, “55% of Companies That Slashed Jobs for AI Now Regret It After Restaffing Costs Soar,” 2026. Reporting on the Careerminds February 2026 survey of 600 HR leaders and Forrester’s Predictions 2026 report.
• Quartz, “Companies Are Rehiring Workers They Replaced With AI After Automation Fell Short,” by Colleen Cabili, July 3, 2026. Reporting on Ford, Commonwealth Bank of Australia, and IBM’s reversals, sourced to CNBC.
• Metaintro, “The AI Boomerang: Why Companies Are Quietly Re-Hiring the Workers They Just Fired,” 2026. Reporting on Robert Half rehiring data and Klarna’s reversal and 2026 breakeven.
• Yahoo Finance / Axios, “‘Job Hugging’ Is on the Rise as Workers Cling to Jobs,” 2026. Reporting on the ResumeBuilder.com February 2026 survey of 2,188 U.S. workers.
• Founder Reports, “Job Hugging Statistics: 2026 Data on Why Workers Are Staying Put,” 2026. Data from MetLife’s 2026 Employee Benefit Trends Study.
• Adamarie, “One-Way Doors and Two-Way Doors: A Simple Framework for Hard Career Decisions,” 2026. Application of the reversible-decision framework popularized by Amazon to individual career choices.

I use AI tools to help research sources and draft these posts. I review, edit, and take responsibility for everything published here, including any errors.