Remember ATMs?
When automatic teller machines rolled out in the 1970s and 80s, everyone assumed bank tellers were toast. The machines could dispense cash, check balances, and handle transfers. Why would you need a human for that?
But a funny thing happened. The number of bank tellers in the US actually increased after ATMs spread. Between 1970 and 2010, the number of bank tellers grew.
How? Because ATMs made it cheaper to operate a bank branch (fewer tellers needed per branch), so banks opened more branches. And the remaining tellers shifted from cash-handling to relationship-building — helping customers with loans, investments, and complex transactions.
The technology didn't eliminate the job. It transformed it. And the transformed version was actually more valuable.
I keep this story in my back pocket because the AI version is playing out right now, and almost everyone is misreading it.
The Paradox Nobody Can Explain
Here's the data that confuses everyone — including many economists.
Multiple validated frameworks have measured AI "exposure" across hundreds of occupations. The numbers are significant: by various estimates, 19-46% of workers could see meaningful portions of their tasks affected by AI. Big, scary numbers.
But when researchers at Yale's Budget Lab and the St. Louis Federal Reserve went looking for the employment consequences of this exposure, they found... not much.
The correlation between AI exposure and actual employment changes is remarkably weak — about 0.47, which in research terms means "there's something there, but it's far from deterministic."
In plain English: knowing how exposed a job is to AI tells you almost nothing about whether people in that job are actually losing work.
How is this possible? How can nearly half the workforce be "exposed" to AI but employment be broadly stable?
Three reasons. And understanding them is crucial for anyone trying to plan their career.
Reason 1: Exposure ≠ Adoption
This is the big one.
AI can write legal briefs. But most law firms haven't restructured their workflows to incorporate AI-drafted briefs into their production process. AI can analyze financial statements. But most accounting firms still have junior accountants doing the line-by-line work.
Only 1% of companies consider themselves "mature" in AI deployment. Almost two-thirds have no formal AI strategy at all.
The gap between "AI can technically do this" and "organizations have actually deployed AI to do this" is enormous. It's like saying "cars can technically go 200 km/h" and concluding that everyone's commute is now twice as fast. The technical capability exists. The infrastructure, culture, regulation, and organizational readiness to use it? Mostly doesn't.
This is genuinely good news for workers — but with a major asterisk. The asterisk is that adoption is accelerating. The gap is closing. And when it closes in a specific industry or function, the change can be abrupt.
Reason 2: Complementarity Changes Everything
This is Daron Acemoglu's big contribution, and it's arguably the most important insight in AI labor economics.
When AI enters a workplace, the outcome depends entirely on how it's deployed:
Automation = AI replaces human tasks. The work gets done by machines, and fewer humans are needed. This is the dystopian scenario everyone fears.
Augmentation = AI enhances human capabilities. Humans do the same work but better, faster, or at higher quality. This often creates more demand for the human, not less.
Acemoglu's research suggests we're currently tilted too far toward automation and not enough toward augmentation. His estimate: AI will increase US GDP by 1.1-1.6% over a decade. Meaningful but modest — and far below what optimists promise.
But the breakdown matters enormously. In roles where AI augments (management consulting, strategic analysis, creative direction), employment is stable or growing. In roles where AI automates (data entry, basic coding, routine customer service), employment is declining.
Same technology. Opposite outcomes. The variable isn't the AI — it's how it's used.
Reason 3: New Tasks Emerge
Here's something the doom-and-gloom crowd consistently misses: when AI handles old tasks, new tasks appear.
This isn't magical thinking — it's documented history. Every major automation technology has created new categories of human work that didn't exist before.
ATMs created personal banking advisors. Spreadsheets created financial analysts. The internet created social media managers, UX designers, data scientists, and content strategists. None of these roles existed before the technology that "threatened" their predecessors.
We're already seeing this with AI. New roles emerging include: AI orchestration specialists, prompt designers, AI ethics consultants, human-AI workflow architects, AI training data curators, and AI output quality auditors.
More broadly, as AI handles routine cognitive work, human attention shifts to work that AI can't do: creative synthesis, relationship management, strategic judgment, ethical reasoning, and complex problem-solving in novel contexts.
The net effect isn't fewer jobs — it's different jobs.
Where It IS Hurting (And Why You Should Pay Attention)
None of this means AI is harmless. The pain is real — it's just concentrated in specific places that the "average employment" numbers miss.
Entry-level workers are getting crushed. Early-career workers (ages 22-25) in AI-exposed occupations have seen 13% employment decline relative to less-exposed occupations. Entry-level job postings dropped 15% year-over-year. This is the real crisis — and it deserves its own article — because AI is eliminating the ramp that junior workers used to develop into senior workers.
Specific task categories are being hollowed out. Not entire jobs, but specific tasks within jobs are disappearing. The legal researcher who spent 60% of their time on case law research. The analyst who spent 70% of their time pulling and cleaning data. The marketer who spent half their time drafting first versions of copy. The tasks are going away even if the jobs persist.
Speed of change varies wildly by firm. Two accountants in different firms face completely different realities. If Firm A has aggressively adopted AI and restructured workflows, accountants there are already working differently. If Firm B is still running on 2019 processes, those accountants feel no impact yet — but they're about to.
The Bottom Line
The job you thought would disappear may actually grow. But it won't grow by staying the same. It'll grow by transforming — by shifting from the tasks AI can handle to the tasks AI can't.
The bank teller who counts cash is gone. The bank teller who builds client relationships is thriving.
The financial analyst who pulls data and makes charts is disappearing. The financial analyst who interprets what the data means and advises clients on what to do is more valuable than ever.
The question isn't "will my job survive AI?" The question is: "Am I building the skills that the transformed version of my job requires?"
Because the transformed version is probably more interesting, more strategic, and more valuable than the current version. But it requires different skills — judgment instead of execution, orchestration instead of production, relationships instead of reports.
The job market isn't shrinking. It's shapeshifting.
Make sure you're shifting with it.