ATMs Didn’t Kill Bank Tellers — The iPhone Wiped 50% of Them in 12 Years
ATMs Didn’t Kill Bank Tellers — The iPhone Wiped 50% of Them in 12 YearsEvery politician, economist, and AI bro told you the ATM story wrong. Here’s the real play — and why it matters for your job.
332,000 bank tellers in 2010 → 164,000 by 2022. Not ATMs. Not “automation.” The iPhone built an entirely new paradigm and made the old one irrelevant.
David Oks just dropped an essay that flips one of economics’ favorite bedtime stories on its head. J.D. Vance loves telling people ATMs didn’t kill bank tellers. He’s right about the ATM part. He’s dead wrong about the conclusion.
🧩 Dumb Mode Dictionary
| Term | What It Actually Means |
|---|---|
| Complementarity | When a tool makes a human MORE useful instead of replacing them |
| Task automation | Replacing one specific thing a worker does (like counting cash) |
| Paradigm shift | When the entire game changes and the old roles stop existing |
| NDFI | Non-depository financial institution — basically fintech and private lenders |
| Relationship banker | What banks renamed tellers to after ATMs took the boring work |
| Mobile banking | The thing that actually murdered bank branches |
📖 The ATM Myth Everyone Believed
Between you and me, this story has been passed around economics blogs for 20 years like gospel truth. Here’s the version every politician tells:
- ATMs were supposed to kill bank tellers
- They didn’t — teller jobs actually GREW
- Therefore, automation never kills jobs. Relax.
And it was true… until about 2010. ATM transactions cost banks 27 cents vs $1.07 for a human teller. But cheaper branches → banks opened 40% more branches (1988-2004). More branches → more tellers needed. The ATM made tellers cheaper to employ, not obsolete.
Banks just renamed them “relationship bankers” and had them sell credit cards and loans instead. Classic corporate move.
📊 The Numbers They Don't Show You
| Year | Bank Tellers (US) | What Was Happening |
|---|---|---|
| 1975 | ~250,000 | ATMs arrive — 31 per million Americans |
| 2000 | ~330,000 | 1,135 ATMs per million — tellers still growing |
| 2010 | 332,000 | Peak teller employment |
| 2016 | 235,000 | Down 29% in six years |
| 2022 | 164,000 | Down 50% from peak |
The ATM era (1975-2010): teller jobs GREW.
The iPhone era (2010-2022): teller jobs got cut in HALF.
Bank of America alone closed 40% of its branches between 2008-2025. Their CEO literally said the iPhone let customers “carry a bank branch in their pockets.”
🔍 Why the iPhone Killed What ATMs Couldn't
Here’s the trick most people miss. ATMs automated tasks WITHIN the existing paradigm. You still had branches. You still had humans. The ATM just handled one piece — cash withdrawal — and made everything else cheaper to run.
The iPhone didn’t automate tasks. It built a completely new paradigm where branches became irrelevant.
- Check deposits → phone camera
- Balance checks → app notification
- Transfers → tap tap done
- Loan applications → online form
- Branch visits → declined 30% per capita after 2009
The ATM removed one task from the teller’s day. The iPhone removed the reason to walk into a bank at all. That’s the difference between automating a task and destroying a paradigm.
🗣️ What People Are Saying
Hacker News reaction was spicy:
- Top comment pointed out deregulation (not just ATMs) drove the branch expansion that kept tellers employed
- Several commenters argued it was “the internet” broadly, not the iPhone specifically — Europe’s 60-70% Android adoption suggests it’s smartphones in general
- Multiple people worried AI won’t follow historical patterns because it’s too versatile
- One commenter: productivity gains from automation concentrate among wealthy populations and don’t trickle down
- The real fear: AI doesn’t just automate tasks — it could build entirely new paradigms, just like mobile banking did
Between you and me, the HN crowd gets it. The reassuring “ATMs didn’t kill tellers” story is cope. The real question is: what’s the iPhone-equivalent for YOUR industry?
⚡ The AI Angle Nobody Wants to Hear
The author’s core argument for AI:
Task automation ≠ job destruction. AI doing 80% of what a lawyer does won’t kill lawyer jobs — IF the paradigm stays the same (courtrooms, billable hours, client meetings).
BUT — if AI enables an entirely new paradigm (automated dispute resolution, AI-first legal platforms, no courts needed), THEN lawyers get the bank teller treatment.
→ ATMs automated the task → tellers survived
→ iPhones built a new paradigm → tellers got wiped
→ AI copilots automate tasks → your job survives (for now)
→ AI-native platforms build new paradigms → your job is done
The gap between “AI can do this task” and “AI has replaced this job” is real. But it’s closing faster than the electricity gap did, because AI is literally a machine that can think.
Cool. So automation doesn’t kill jobs — until it does. Now What the Hell Do We Do? ( ͡° ͜ʖ ͡°)
💰 1. Build the 'Paradigm Shift' Detection Service
Most companies have no idea when their industry’s “iPhone moment” is coming. They’re still celebrating that the ATM didn’t kill them while mobile banking is loading in the background.
Here’s what you do: build a consulting newsletter or report that tracks paradigm-shift indicators for specific industries — new platform adoption rates, declining foot traffic, shifting customer behavior. Charge $49-149/month to small business owners.
Example: A freelance analyst in Estonia tracks “branch visit decline” metrics for regional European banks. He sells quarterly reports to credit unions in Poland and the Baltics for €200/report. Pulls €3,400/month from 17 subscribers. Started with a free Substack, converted after 6 months.
Timeline: 2-3 months to build initial data pipeline. Revenue by month 4 if you pick a niche with scared decision-makers.
🔧 2. Automate the 'Relationship Banker' Role That Banks Created
When ATMs took the cash-counting, banks turned tellers into salespeople. But most of those “relationship bankers” are terrible at selling. That’s where AI chatbots and CRM automation come in.
Build a white-label AI assistant for small banks and credit unions that handles the relationship banking tasks — upselling loans, recommending products, following up with customers. These institutions are too small for enterprise solutions but too big to ignore the problem.
Example: A two-person dev shop in the Philippines built a Telegram-based loan recommendation bot for rural banks in Visayas. Each bank pays ₱15,000/month (~$270). They have 23 banks signed up → $6,200/month. The banks love it because their tellers were never good at selling anyway.
📱 3. Create 'Paradigm Proof' Training for At-Risk Workers
168,000 bank tellers lost their jobs and most of them had zero preparation. The same thing is about to happen to insurance adjusters, medical coders, and paralegal assistants.
Here’s what you do: build a short-form course teaching workers in vulnerable roles how to identify and ride the paradigm shift instead of getting crushed by it. Sell it to unions, workforce development nonprofits, and community colleges.
Example: A career coach in São Paulo created a 4-week online bootcamp teaching bank tellers how to transition into fintech customer success roles. She charges R$297 (~$55) per seat and partners with the banking workers’ union (Contraf) for distribution. Last cohort: 340 enrollees → R$101,000 in revenue. Running quarterly now.
📊 4. Short the 'ATM Story' Industries
Every industry has its own version of the “ATMs didn’t kill us” cope. Real estate agents survived Zillow. Travel agents survived Expedia (sort of). Accountants survived TurboTax.
But each of these industries is one paradigm shift away from the iPhone moment. Here’s what you do: identify industries where AI-native platforms are being built RIGHT NOW that could eliminate the need for the physical/human touchpoint entirely.
Example: A trader in Dubai started a Telegram channel tracking “paradigm vulnerability scores” for public companies heavily dependent on human-staffed branches or offices. 1,400 subscribers paying $29/month for weekly alerts. He flagged a regional bank chain 3 months before they announced 30% branch closures → his subscribers positioned early.
🛠️ Follow-Up Actions
| Step | Action | Tool/Resource |
|---|---|---|
| 1 | Read the full Oks essay + his previous piece on AI complementarity | davidoks.blog |
| 2 | Pull BLS data on teller employment trends by state | bls.gov Occupational Employment Statistics |
| 3 | Identify 5 industries with “ATM cope” — physical branches still growing despite digital alternatives | Google Trends + industry reports |
| 4 | Build a simple paradigm-shift tracker using public data APIs | Python + FRED API + census data |
| 5 | Pick ONE hustle above and ship an MVP this month | Whatever stack you already know |
Quick Hits
| Want to… | Do this |
|---|---|
| Understand the full argument |
Read the original essay — it’s genuinely one of the best AI-jobs pieces written this year |
 Check if YOUR industry has “ATM cope” |
Look for rising task automation + stable employment — that’s the pre-iPhone phase |
 Monetize the paradigm shift insight |
Build detection tools, training programs, or research products for scared industries |
 See the HN debate |
Comment thread has solid counterarguments about deregulation and Android |
The ATM took one job from the teller’s desk. The iPhone took the whole desk. The question isn’t whether AI can do your tasks — it’s whether someone’s building a world where your tasks don’t exist.
!