Federal Reserve Research Shows AI Is Creating Two Very Different Job Markets, Divided by Experience

Dallas Fed: AI is replacing entry-level workers and rewarding experienced ones

Today’s job market headlines include a new report that shows the U.S. labor market just hit its lowest point for worker turnover in nearly a decade. And a major policy paper from Brookings was just published. It reveals a massive gap in the U.S. tax code that may be quietly subsidizing automation over American workers. It’s a really interesting angle, one I hadn’t previously thought about or seen mentioned anywhere. So I look forward to getting into all of that. But first, the Federal Reserve Bank of Dallas just published a piece that I think is one of the most important things written about AI and jobs this year. Since ChatGPT launched in the fall of 2020, so just past three years, total U.S. employment has grown about 2.5%. But the computer systems design sector, which is one of the most AI-exposed industries of all, it saw an employment drop of 5%. And across the top 10% of AI-exposed industries overall, employment declined about 1%. So AI-exposed jobs have been going in the wrong direction.

But across the entire US, average weekly wages rose 7.5% since 2022. And in computer systems design, they’re up 16.7%, which is more than double the national rate. The top 10% of all AI-exposed industries saw an 8.5% wage growth overall. So while there’s fewer jobs, the people who are able to keep them, they’re now earning significantly more. And that seems like a contradiction, but it’s not. It’s a pattern. He’s an assistant vice president in the research department of the Dallas Fed. He refers to it as an experience premium. That’s the wage gap between experienced and entry-level workers in a given occupation. The median experience premium across occupations sits at about 40%, but the range is massive. Fast food workers and ticket agents or low-level jobs, the difference is only about 10%, actually less.

But lawyers, insurance underwriters, credit analysts, people who are specialized, the difference is over 100%. The key finding is that occupations with higher experience premiums have a stronger positive relationship between AI exposure and wage growth. In plain language, that means AI is a tailwind for experienced workers in knowledge-heavy fields and a headwind for entry-level workers in those same areas. As Davis puts it, AI can substitute for entry-level workers and at the same time complement experienced workers. The explanation comes down to two types of knowledge: codifiable knowledge, which is stuff you’d read in a book or learn in school, and tacit knowledge, which is the judgment and intuition that only come from years of hands-on experience. AI is very good at replicating codifiable knowledge, but it can’t replicate tacit knowledge. At least not yet. And for entry-level workers, those codifiable tasks represent the core of what they do day-to-day. So all of this confirms what I’ve seen in other reports, what everyone who’s in the job market has been feeling, what my clients say, what my peers and staffing say, that AI isn’t really replacing experienced professionals at this point. In a way, and this data backs it up, it’s making them more valuable, at least in some cases.

But the pain for entry-level workers, young professionals, people coming out of school, that is very real. We know that that’s happening right now. This data confirms it. So if you are a young professional, you need to build tacit knowledge and real-world experience as fast as you can. You need to figure that out. Maybe it’s internships, mentorships, project-based work, or doing things on your own in your own field. Those opportunities exist. Don’t wait for someone else to hand it to you. Your school’s not going to do it. We know that now. The market’s tough. It’s only going to get worse from everything I’ve seen. And now this is real data that backs it up. So take control of your own destiny if you’re a young professional.

ADP: The labor market just hit its lowest turnover in nine years

So let’s shift from AI to the broader labor market. ADP Research put out a piece this morning that paints a pretty stark picture of where things stand right now, as do all of them lately. Nella Richardson, who is ADP’s chief economist, looked at the labor market through two fundamental economic variables, job quantity and wage growth. And when you put them together, well, the takeaway is pretty sobering. Hiring has slowed sharply over the past three years, and the financial payoff for switching jobs has dropped to its lowest point in ADP’s data going back to 2017. So think about what that means. During the great resignation, the post-COVID days, around late 2021 and 2022, changing jobs meant getting a significant pay bump. It was real money, better benefits, employers were bending over backwards. But that era is clearly over. Employer turnover, which is measured as quits and layoffs as a share of total employment, hit its lowest level in nine years. So employees are not leaving.

They certainly don’t want to leave, nor should they in this market. But all the data is really just pointing in that direction right now. We haven’t seen massive layoffs, but we’re seeing it becoming increasingly harder to get jobs unless you’re a specialist. And that is something that I think is going to be the future of the U.S. labor market. Pay growth is stabilized above pre-COVID levels, but the dynamic push and pull between job gains and pay growth that typically keeps the market healthy, well, that’s weakened considerably. Richardson summed it up by saying the labor market is now defined more by inactivity than vigor. That line says it all.

Brookings: How the U.S. tax code is tilting the playing field toward automation

And then the final story today really ties all this together with an interesting angle. I said that at the beginning, it’s unique. I hadn’t thought of it. But as more companies realize, I don’t think many employers have, this is going to be a factor that makes job displacement significantly worse than it otherwise would be, and limit, or rather, encourage the use of AI. So the Hamilton Project at Brookings is who uh publishes this, and it’s a policy paper. It’s not something I normally talk about, but it looked at how U.S. policy shapes whether AI helps workers or replaces them. So very relevant to the job market. And the core framework is simple but really powerful. AI can be used to augment workers, making them more productive, or it can automate workers. So is it going to help workers or is it going to replace them entirely? And right now, the authors argue market forces in the US policy are tilted heavily towards automation or replacement.

Here’s how it breaks down. The effective marginal tax rate on labor in the US is between 25.5% to 33.5%. While the effective tax rate on capital equipment that’s used for automation is around 5% after depreciation deductions and investment credits. So think about that. It’s costing you, in addition to any other challenges that come with employees, between five and six times more in taxes alone. And as a staff and company owner, I know all too well how expensive it is to uh get hit with payroll taxes from state to state. They vary greatly. But this is a massive gap we’re talking about, no matter what state you’re in, because it essentially means the tax code is subsidizing the replacement of workers with machines. I called it inadvertent at the beginning of this episode. I think it is inadvertent, but it’s also very real. And it’s so real you don’t need an economist to predict what happens once everyone realizes that, right? When costs dramatically, when it costs dramatically less to deploy AI than to employ a human, what do you think businesses are gonna do? We know what they’re gonna do, what they’ve always done. They’re gonna find a way to operate with less expenses, do more with less, and this is a big incentive to not hire people to automate instead.

The paper notes that roughly 80% of US workers could see at least 10% of their tasks affected by large language models. And about 19% of workers may see 50% or more of their tasks impacted. Okay, so they raise that. It’s a range, but I I still think the numbers are bigger. I think 80% of US workers are going to see um 50% or more of their tasks impacted in the relatively near future. So I’m very confident in AI’s ability to continue to evolve and improve rapidly. Um not everyone is, I get it. I hope I’m wrong. I say that often, that’s for sure. But that’s what this uh research report is telling us. There is a positive side to the story. Uh, research cited in the paper found that AI-assisted workers saw productivity gains, annual productivity gains of around 14%. And the least experienced workers, those are the ones that are getting hit the hardest, as I talk about often, as I already talked about today, they saw productivity gains up to 34%.

That’s a big deal. It means AI has the potential to be an equalizer, giving newer workers the tools they need to perform at a higher level faster than they otherwise would. But that’s only if companies deploy it that way. And every company is going to have to make up their own minds on this. I don’t see the tax code changing. The paper uh also references research that shows between 50 and 70 percent of changes in the U.S. wage structure over the past four decades can be attributed to automation displacing workers. And the policy recommendations include reforming the tax code to level the playing field, where they would expand, they recommend expanding RD credits to cover worker augmentation investments and updating safety net programs like unemployment insurance.

Sure. Good luck holding your breath on that one. We can’t even fund the government, we can’t agree on anything right now, and this would take a major shift. It would take a big shift at the state level. If you’re not familiar with labor taxes, payroll taxes from state to state, they vary widely. I mean, wildly. So we’ll see what happens there, but I’m certainly not going to expect that the government is going to come in and change anything anytime soon, unless or until job displacement gets really bad. So I think this paper is recommending take action before that happens. I doubt we’ll see it, but eventually it may be necessary. Companies are going to have to be incented to hire real people instead of letting robots do the work. And right now it’s the opposite. That’s that is very clear after seeing uh the numbers in this paper. So good stuff today.

Fun fact: The surprising upside of cyber slacking

Um, not a huge news day, but some interesting research that gives a different perspective on what’s going on in the market and explains what many of us are feeling and seeing day to day. So that’s it for today. But before we close, here’s your fun fact Cyber Slacking, which is browsing the web at work, supposedly can actually help restore energy if it’s kept at 1015 uh 10 minute burst, right? So if you just do it for a few minutes, it’ll help you restore energy. Yeah, but what if you’re doing it every 10 minutes and then taking a 10 minute burst? It doesn’t say anything about that. So I don’t know. Cyber slacking. I like the term though, I haven’t heard it before. So be sure to implement it. I do my share of cyber slacking during the day, that’s for sure. Uh can’t say off, can’t say off Twitter. Anyway, thanks for listening. I really appreciate it. Please like, subscribe, share with anyone who might be interested, and I look forward to talking to you tomorrow.