Home / Blog / If AI is replacing engineers, why are companies still hiring them?

If AI is replacing engineers, why are companies still hiring them?

You see the take constantly on LinkedIn. "If AI is replacing software engineers, why are companies still hiring them?" It's catchy, probably meant in good humor. But it's also missing the point — and the history.

The ATM story

In December 1973, the New York Times ran a piece titled Machines — The New Bank Tellers, warning that ATMs could displace up to 75% of teller jobs. The machines were coming. The math seemed clear.

Here is what actually happened over the next four decades:

The number of ATMs grew 37-fold between 1975 and 2000. Teller employment doubled in the same period.

The explanation is not complicated once you see it. The number of tellers needed to run an average urban branch fell sharply, from around 20 down to 13. But banks responded to lower operating costs per branch by opening more branches, competing aggressively for market share. Total employment across the industry grew even as the productivity of each branch improved.

More importantly, the job itself transformed. Tellers who used to spend most of their time counting cash started spending more time building customer relationships, handling complex products, and solving problems a machine could not be trained to anticipate. The role became more skilled, not less.

The IMF made the same point in 2015 in James Bessen's piece "Toil and Technology": automation does not equal unemployment. It equals reallocation. What gets reallocated is the work that machines are not good at.

The parallel is hard to miss

The "AI is replacing engineers" framing treats the job as a fixed quantity of tasks, where automation reduces the total. That is not how it has worked historically, and there is no reason to think it will work that way now.

What AI is doing to software engineering looks a lot like what ATMs did to banking. The scaffolding work, the boilerplate, the parts of the job that are mechanical and repeatable, those are the things getting automated first. In the teller analogy, that is the note counting.

What remains is harder to automate and more valuable: understanding what actually needs to be built, designing systems that will hold up under real conditions, making judgment calls when the requirements are contradictory or incomplete, working with people who have different mental models of the problem, and catching the things the AI got confidently wrong.

The CCIA made the case in 2026 using both bank tellers and radiologists as examples. Radiology got AI-assisted diagnosis tools that were supposed to hollow out the profession. What happened instead was that the volume of imaging increased, radiologists spent more time on complex cases that required clinical judgment, and the role evolved around the things the tools could not do reliably.

The pattern holds across domains: automation shifts demand toward higher-skill work, and it often expands the total market for the service in the process.

Why the iPhone matters more than the ATM

The decline in teller employment since 2010 is real, and it is worth being honest about. But as David Oks argues, what drove that decline was not the ATM. It was mobile banking. When the transaction volume that used to require a branch visit moved to a phone, the economic case for maintaining branch density collapsed, and teller employment followed.

That distinction matters. The ATM was a tool that made branch operations more efficient, which expanded the market. Mobile banking was a substitute for the branch itself, which contracted the market.

For AI and software engineering, the question is which of those two dynamics is playing out. Right now, AI coding tools look more like ATMs than like the iPhone. They make engineers more productive on the tasks they are already doing, which tends to expand what teams can build and ship, not shrink the headcount needed to build it.

The risk to watch is the substitute scenario: AI systems that take over whole categories of software that used to require custom development. Internal tooling, simple applications, well-defined automation tasks. If enough of that category moves to "AI just builds it," the downstream effect on hiring could look more like the iPhone than the ATM.

That is not where we are today. But it is where the incentives are pointing, and it is worth being honest about.

What this means for engineers

The teller analogy is not a reason to relax. The tellers who thrived after ATMs were not the ones who kept doing what they had always done. They were the ones who built skills in the parts of the work that the machine could not touch.

The engineers who will be in demand in five years are not the ones who can use AI tools to write code faster. That will be a baseline expectation, not a differentiator. What will matter is the ability to do the work that comes before and after the code: understanding what actually needs to be built, decomposing complex systems, exercising judgment when the spec is ambiguous, and catching problems before they become expensive.

If you are still spending most of your time on the metaphorical note counting, the transition is going to be uncomfortable. Not because AI is replacing engineers. Because the job is transforming, the same way it always has when a new tool arrives and takes over the mechanical parts.

The future is not engineers who can use automation. It is engineers who can do the work automation cannot.