Bank tellers didn't disappear. That's the whole point.
In 1970, there were roughly a few hundred thousand bank teller jobs in the United States. Then automated teller machines rolled out nationwide over the following decades, eventually numbering in the hundreds of thousands, each one able to do the core cash-handling task of a teller in seconds, without a lunch break. If you'd asked people in 1970 what happens to teller employment once machines can dispense cash, most would have guessed it collapses. It didn't. By the 2000s there were actually more bank tellers employed in the US than before ATMs existed, even after adjusting for population growth the decline was far more gradual than the raw automation math suggested.
What happened instead is a genuinely interesting economic story, and it's real, documented, and taught in labor economics courses: ATMs made each branch cheaper to run, so banks opened more branches, and tellers whose core cash-dispensing task got automated shifted toward the parts of the job machines couldn't do — building relationships, selling loans and accounts, handling disputes, and solving problems that needed judgment. The job didn't vanish. It got recomposed around a different mix of tasks. This is the single most useful mental model for thinking about AI and employment today, and it's the reason economists keep bringing it up.