AI and the Decoupling of Jobs from Economic Growth

What if working fewer hours coincided with stronger economic growth?

New labor data reveals a puzzling trend: employment is being revised downward even as the economy keeps growing. Professor Harvey explores whether AI is the variable that can help us understand that divergence.

The conversation moves beyond the familiar narrative of job displacement to examine a structural shift in how AI functions. As systems evolve from tools into agents, they can organize tasks, execute workflows, and contribute to their own development.

As structured, repeatable white-collar work becomes increasingly automated, productivity gains may allow output to grow with fewer hours worked per person. The result may be a reallocation of labor rather than a simple reduction, with potential implications for economic growth and fiscal capacity.

Transcript

SPEAKER_01 0:00

Thank you once again for joining us for our time together on uh Cam Harvey's Through the Noise. Cam, we've revisited AI through a couple of our of our sessions. And one of the narratives around AI is that it's it's gonna grab all the jobs and and lead to uh to some bad outcomes. And and recently we've noticed negative growth in the job market. We've seen negative revisions in the in the data that's coming out. And I thought maybe we should spend a little bit of time digging into that.

SPEAKER_00 0:34

Sure. Um this is an extraordinary time, and I believe that February 5th, 2026 will go down in history as a historic day. Why is that? So that was the day that Anthropics Claude 4.6 was released. And you might say, well, it's no big deal, 4.5, 4.6. No. The news was not the release of 4.6, the news was that it was 100% coded by AI. So instead of like a team of people doing the programming, it was AI coding a version of AI. And this is called in computer science recursive self-improvement. And it is a very, very big deal. So, rightfully so, people took that and saw the potential of a substantial disruption in in the job market uh for many white-collar jobs. So this was, again, uh February 5th is is that point very significant.

SPEAKER_01 1:54

Aaron Powell So you've seen um you've seen this recursive thing happens when I'm looking at the the jobs out in the marketplace. Which ones do you think are most at risk?

SPEAKER_00 2:06

Yeah, so uh so let me kind of step back a little bit before I get to the winners and losers. So there are many uh jobs that will be replaced, but we're seeing this just in general. So you mentioned at the beginning uh a revision. Um so uh we had a revision where we lost one million jobs, and and nobody noticed. The GDP is growing at a reasonable rate, and and we lose a million jobs, and it doesn't really affect anything. And that's because of productivity gains with the existing AI that we already have. Another example is uh Jack Dorsey's block, they lay off 4,000 people, the stock price goes up by 24%. So it that making an announcement that we're going to pivot uh towards AI, and I think also the block story is more complicated, they might have overhired in the first place, but nevertheless, that's an important signal to the market. So the point is that this technology allows uh people to be much more productive than in the past, and that means that some jobs will be replaced, and those that remain become dramatically more productive. And and when you put that together, you get GDP growth. So you would think that with employment going down, that's usually a negative association with growth. That's not necessarily the case in this particular situation if we've got uh if we've got this pickup in productivity. And then this is pretty important. Uh I talk to my colleagues uh about this all the time, and some of my colleagues will refer to AI as a tool, and it is no longer a tool. Uh a calculator is a tool. So we're moving to a world where the AI is an agent. And that's much different. So the agent can organize, make decisions, execute a plan. So this is totally different than anything that we have seen before.

SPEAKER_01 4:34

When you say agent, you mean that it has some autonomy that it can make, I don't want to say decisions because we're putting a human trait on these things, but it but it can act without your direct input like a tool. If you allow direct input.

SPEAKER_00 4:49

Yeah, if you allow it. And and that's essentially the announcement on February 5th that, oh, well, this version, um, the Claude 4.6 was coded by AI. So so this is a big deal, and it's something that we've not seen historically. We've got just unprecedented uh computing power, and there's no sign that that will be uh diminishing. For me, it is kind of remarkable seeing Nvidia's Hopper, Blackwell, the Rubin that they've announced. That in my time there was a supercomputer called the Cray XP2, and it was remarkable. It did 1.6 trillion operations per second. Well, the latest NVIDIA chip is 25 million times as powerful as that. So again, this conflux of uh of software and computing power ushers in a new age, and there will be winners and losers. And most of the conversation has been on the losers. And let me unpack uh just a few. So think of tax preparation as we approach uh April 15th. That's just an algorithm. The tax code is complex, and we employ uh tax accountants or we buy software to do this. In the future, that will be done by AI for your$20 a month uh subscription, just part of the package. And within accounting and tax preparation, there's maybe three million jobs. And if we assume half are replaced, that's 1.5 a million. Another interesting uh field is law. So again, law is it's not as algorithmic as tax preparation, but nevertheless, the large language models are really good at the law. And if you look at the total number of lawyers, it's about 900,000 in the US, and then maybe 300,000 paralegals. And then every year, 36,000 new lawyers graduate in the US. I'm thinking half. So we're gonna lose half. So again, junior lawyers coming in, searching for precedent for a particular case, that could be done instantly with AI. So there will be very significant and fast disruption. So is this a disaster for the U.S.? I don't think so. So um I've talked to you about this before, that fortunately the US is the leader in AI right now and other disruptive uh technologies, a very rich uh country. And uh and many of those that are disrupted will be reallocated to more productive uh uses. Uh there will be a window, perhaps, of a couple of years where it will be difficult, but I do believe they will be relocated, and some will elect to work fewer hours to make room uh for more people. So think of some people choosing to do 20 hours rather than 35. Uh, so that's like two jobs that one person did uh in the past. So I don't see this as a mass unemployment situation. I see this as a reallocation of human capital in a way that uh is fueled by productivity growth, more leisure time, and uh the standard of living actually goes up. So I think that this holds a lot of promise for economic growth, and growth is something that our economy needs, especially given the size of the US debt at$38 uh trillion dollars. We need growth. So the growth is ideal because if you grow, then the tax revenue just naturally increases, and we need that. And I do believe it's feasible to move away from this oh, well, 1.9% growth is good, to we want 5 to 7% real GDP growth. And that can make a dent uh in the deficit and our debt.

SPEAKER_01 9:50

Well, this is a fascinating take on this disruptive um technology that we're all we're all experiencing it on a daily basis, and it's changing so rapidly. So um it will be interesting to see how all of this plays out, Cam. Definitely.