AI is freezing the first rung. Support is where it starts.

There is a number worth sitting with. Stanford’s Erik Brynjolfsson found that workers aged 22 to 25 in the roles most exposed to AI, software development and customer service among them, have seen a double-digit percentage drop in employment since 2022. Not because they were fired in a wave. Because they were never hired in the first place.

That distinction matters more than the headlines suggest. Most coverage of AI and jobs still reaches for the layoff story: a company announces a cut, blames automation, moves on. And those cuts are real. Challenger, Gray & Christmas counted nearly 50,000 job cuts tied to AI so far this year. But the layoffs are the loud, visible part. The quieter thing is a hiring freeze at the bottom of the ladder, and it is doing more damage than the announcements.

A freeze, not a purge

Look at where the pain actually lands. The unemployment rate for Americans aged 22 to 27 has climbed to just under 6 percent, roughly a third higher than the national average and close to double where it sat before the pandemic. One university study found that headcounts of junior employees fell after early 2023 while the ranks of senior staff held steady. Nobody got marched out. The door just stopped opening.

Customer service is squarely in the blast radius. It has always been one of the classic first jobs, the place a lot of people start, and it is exactly the kind of routine, high-volume work generative AI handles well. So the postings thin out. A company that used to bring on five new support reps a year brings on one, or none, and lets a bot take the overflow. On a spreadsheet this looks like pure efficiency. The salary line drops and the tickets still get answered.

The spreadsheet is missing something.

What the first rung was actually for

Entry-level support was never just cheap labor answering easy questions. It was where a company learned about itself. The new rep who handles forty tickets a day finds out which product confuses people, which policy generates the most anger, which workaround the docs never mention. That knowledge accumulates in a human being, and over a couple of years that human becomes the team lead who trains the next batch, then the ops manager who redesigns the broken policy.

Cut the rung and you do not just save a salary. You cut the pipeline. As one analyst put it plainly on a recent industry podcast, freezing junior hiring saves money in the short term but leaves you with “a weaker future pipeline for middle management” and less institutional knowledge across the org, which means leaning on expensive outside hires later. Forbes contributor Bernard Marr made the same point from the other side: entry-level work is where people “learn how the business works, how decisions are made, how expertise is built.” It is the training ground disguised as a cost center.

So a company that automates its whole front line without thinking about this is trading a known, small expense for an invisible, compounding one. It keeps the tickets answered and quietly stops learning about its own customers.

The routine part and the knowledge part are separable

Here is the thing the all-or-nothing framing gets wrong. Automating support does not have to mean torching the institutional knowledge that support produced. Those are two different problems, and you can solve the first without causing the second, if the tool is built for it.

That is the bet we made with IMCeleste. Our agent, Celeste, does not arrive as a generic bot pointed at your customers. She learns from your own support history first, the accumulated judgment your team already built, and she practices privately before she ever goes live. She goes live one topic at a time, and only when the owner decides she has earned it. When she hits something she does not understand, that gap gets surfaced instead of buried, so the knowledge keeps growing rather than draining out the door with the reps you did not backfill. The point is to automate the routine part of the first rung without losing the part that was actually valuable. You can see how the whole model fits together if you want the detail.

The part nobody offers

There is a second, harder thing the current debate keeps circling. On that same podcast, one analyst named the real source of the anger: CEOs keep saying AI will take the jobs, and then there is no follow-up. No quid pro quo. Just an open-ended “your work is being automated, and now what.” That silence is why 70 percent of people tell pollsters they expect AI to shrink job opportunities, and why the backlash is not a flash in the pan.

We do not think a support platform fixes the macroeconomy. But we are not comfortable pocketing the entire gain from work that used to belong to people, either. That is the whole reason for The Dividend Standard, our commitment to direct the greater of 5 percent of revenue or 70 percent of profit to the workers whose jobs this kind of software automates. It is not charity and it is not a press release. It is an attempt to answer the “and now what” honestly, with a number attached.

The first rung is freezing. You can pretend the knowledge it produced was never worth anything, or you can build the automation so that knowledge survives the transition, and share back some of what the transition earns. We think the second path is the only one worth taking. If that is the kind of AI customer service you want to run, come talk to us.