This spring’s labor market statistics don’t exactly reflect the sentiment. You would expect the unemployment line to be wrapped around the block if you were to walk through any tech conference or browse through any LinkedIn feed. Instead, law firms are hiring more people. Seats on the consulting bench are filling up once more. There are bidding wars among accountants. Something isn’t adding up, and one of the year’s most intriguing mysteries is the discrepancy between what people fear and what the data indicates.
The fear is familiar by now. Lawyers, accountants, consultants, marketers, junior analysts, half of middle management, and most likely the person who keeps the office snack drawer organized will all be replaced by AI. Even the white-collar tasks that the models can already perform were mapped out in a recent Anthropic study.
| Topic Snapshot | Details |
|---|---|
| Subject | The Automation Paradox in the AI economy |
| Core Idea | Productivity gains from AI are expanding, not shrinking, white-collar headcount |
| Economic Theory | Jevons Paradox, applied to labor by economist Torsten Slok |
| Affected Sectors | Law, consulting, banking, accounting, software, design |
| Reshaped US Jobs (2–3 yrs) | Roughly 50–55% of roles, per BCG modeling |
| Potential Job Losses (5+ yrs) | 10–15% of US jobs |
| Notable Companies Cited | JPMorgan Chase, Toyota, Inditex (Zara) |
| Counter-Indicator | Youth unemployment fell from 9.2% (Sep) to 5.6% in March |
| Time Horizon | Short-to-medium term; long-term remains uncertain |
However, when one looks at the actual hiring figures, a different picture becomes apparent. In reference to the English economist of the 19th century who observed that improved steam engines caused Britain to burn more coal rather than less, Torsten Slok, chief economist at Apollo Global Management, recently dubbed it the Jevons employment effect. According to Slok, cheaper inputs increase rather than decrease markets. More legal memos result from cheaper legal memos. There are more financial models when they are less expensive. Strangely, more work also equates to more workers. It’s a neat theory. It’s another matter entirely if it holds.
Observing this unfold, it seems that those who are most certain about AI’s ability to eliminate jobs are the ones who are most removed from the actual production floor. Forty years ago, when everyone in Detroit was racing toward what they called “lights-out factories”—plants with no people and only machines humming in the dark—Toyota realized this.
The engineers at Toyota declined. They thought that automation without comprehension resulted in brittle systems. In 2012, the company reversed course and reinstated workers on a portion of its assembly line. Defect rates decreased. Expenses came next. They are currently about 30% more productive than their fully automated rivals. One quote from Mitsuru Kawai, the former Toyota executive who oversaw a lot of this, stuck with me: “People are good at learning, machines are good at repetition.”
The same strategy is being used by banks, but they are dressed more elegantly. No associate could match the speed at which JPMorgan’s COIN system scans credit agreements—roughly 360,000 hours of legal review annually. However, nothing is decided by the system. It gets ready. Signing off is still done by humans. Instead of shrinking, the legal department’s workforce was moved up the value chain. According to the majority of internal accounts, there were 12% cost savings, no layoffs, and an improvement in morale. Automation is most effective when it targets tasks rather than people, a lesson that keeps coming up in various industries with disparate vocabularies.
Zara recounts the same tale in a different way. Its parent company, Inditex, purposefully reserves roughly half of its production capacity in order to respond to what is truly selling. Zara automates the feedback loop—the part that says, “Change course now”—while rivals automate forecasting months in advance. The business reported an operating margin of 17% last year, which was nearly twice the industry average. Automation didn’t speed up Zara. Learning did.
Nevertheless, the optimistic version of this story has flaws that are difficult to ignore. Over time, ATMs did not increase the number of bank tellers. While the accounting industry as a whole expanded thanks to QuickBooks, fewer, more qualified individuals benefited from this expansion. The Jevons paradox can be harsh on an individual basis but true overall. A junior associate may find cold solace in the fact that her company has hired two new partners this quarter.
For the time being, it appears evident that the doomsday timeline keeps slipping. After rising through 2023 and into the fall, the unemployment rate for people aged 20 to 24 has actually decreased to 5.6%. Businesses that make excessive cuts are already discreetly rehiring. There are many companies that automated their way into chaos, and they are paying consultants to fix their mistakes. Nobody really knows if this is the beginning of a longer, more bizarre boom or the calm before a deeper restructuring. Perhaps we are witnessing the early stages of a true expansion. We might also be witnessing the final quiet moment before the floor shifts. In any case, not everyone will be affected by the robots. Not just yet. Someone still needs to give them instructions in the interim.