AI Anxiety Part 2: The ATM Paradox

The paradox

When ATMs rolled out across the U.S. — from almost none in 1970 to hundreds of thousands over time — everyone assumed bank teller jobs were finished. They weren't.

Teller employment went from about 300,000 in 1970 to nearly 600,000 by 2010.

Here's why

ATMs reduced the number of tellers each branch needed — from 21 to 13 between 1988 and 2004. But that also made branches much cheaper to operate, so banks opened more of them. In urban areas, the number of branches grew by 43%.

Fewer tellers per branch × a lot more branches = more tellers overall.

And the job itself changed.

Less cash handling. More customer support, relationship banking, financial products, and problem-solving. The parts of the role that required a human became the valuable part.

The AI parallel

I see the same dynamic with AI and design right now. The mechanical output gets cheaper. The strategic thinking, judgment, and messy human coordination take up more of your day.

The second half

But here's the second half of the story.

After 2010, it reversed. Teller jobs fell from nearly 600,000 to about 347,000. Job postings dropped too.

It wasn't ATMs that finally broke the model. It was mobile banking. Once banking moved onto smartphones, technology could handle far more of what tellers still did — and the role came under much greater pressure.

The real framework

When tech automates some of your tasks, the rest of your work often becomes more valuable.

When tech automates most of your tasks, that stops working.

In design and development right now, we're still mostly in the "some tasks" phase. AI can generate variations, boilerplate, and first drafts. It still can't sit in a room with three stakeholders who disagree about scope and figure out what actually needs to be built.

Not yet.

But "not yet" is doing a lot of work in that sentence.

I don't think the move is to relax because the current moment still looks okay. The move is to keep growing the parts of your work AI can't easily replace — so that when capabilities jump again, that's where you're standing.

Next in series — Part 3: Photography, desktop publishing, Canva, and now Claude — every time someone said design was dead.