The goal of productivity enhancement is going to produce value, justification, only if you either generate more business or fulfill existing business opportunity at a lower cost. The former demands an elastic market for the product/service, and I don’t have enough information to assess how likely that’s the case in the US or global economies, nor do I think there’s any credible source that does. That means that the value of productivity improvement lies in its ability to reduce labor cost, either by reducing labor or by using cheaper labor. That, in turn, means those two factors have to justify any tech project, including AI.
This means it’s important when complications arise in those productivity factors. A recent analysis by Axios may be exposing an issue that faces AI and other cost-management strategies. Productivity tools have historically been aimed at office workers for the simple reasons that the “information content” of their jobs was high, they already had devices to tap into information, and they were conveniently located in fixed positions. As the piece points out, over the last three years or so, US employment has shifted a bit away from office workers. Are the people we’ve depended on for productivity enhancement dwindling in the workforce? If so, why. If so, what’s next for productivity? My view, based on enterprise comments, has been that office workers represent 60% of the workforce, and non-offce, labor, workers the other 40% If we used up the former, we’d have to depend on the latter. OK, but does the available labor statistical data validate, enhance, or repudiate that?
To start with, it’s a bit difficult to validate the Axios data because the categories they define aren’t explicitly in the BLS (US Bureau of Labor Statistics) data. I also distrust the 2025 data because of the combination of government shutdown and politics. Finally, I don’t think that just grabbing recent years gives us a valid trend. So I’ve done my own analysis of the raw BLS information, and I’ll try to address the questions based on this analysis.
There are 21 major BLS occupation categories plus a total-jobs category. I’ve taken enterprise comments to assign them to two groups, “office/desk” jobs that represent people who are largely information workers, and “labor” jobs that involve manipulating real-world things. The former group has 11 job categories, the latter has 10, and I’ve looked at all these and the total employment over a ten-year period.
From 2015 to 2024 (a decade), US employment overall has grown at a CAGR of 1.12%. Obviously, this has to miss the large number of under-the-table workers, most of whom are involved in various personal-service types of jobs, “labor” in a broad sense. Despite this, if we look at the balance of “office/desk” jobs and “labor”, it shifts by only about 2% (58/42% in 2015 to 56/44% in 2024) over the decade. Enterprise views, then, are close to the statistics, but not spot on.
Statics help us dig deeper, in any event. The biggest sources of growth the data shows were in Management, Business/Financial operations, Healthcare Support, and Transportation/Material Moving. Two in each in office/desk and labor. In the same period, the biggest areas of job loss were in Personal Care and Service. There were a total of three labor occupation categories that showed a loss of jobs and two office categories. Two office categories had substantially neutral job growth, compared to three labor categories. So far, not a major difference.
The picture this seems to be creating is driven by “automation” and company revenue opportunity. We’re seeing growth in job categories where the vertical they represent is strong, so the companies can earn a good return on their labor cost. We see slower growth, no growth, and even declines in jobs that are readily automated or that are changed by automation, like sales (where jobs decline because of online shopping). In production occupations, we see a decline in jobs reflecting increased process automation, where in transportation and material-moving jobs that have not been highly targeted by process automation, there’s a significant increase.
But automation hasn’t created revolutionary change, and in fact the end result might validate Axios’ view. If we look at the CAGR for office jobs versus labor, we do find what the Axios piece suggests; the former grew at only 77% of the labor force growth overall, while the latter grew 129%, almost double the rate. However, this has to be explored in two further ways—in the context of the overall economic situation, and by specific job classification—to yield a final result.
There’s a broad indication of a growing sense of financial insecurity among consumers, the ultimate economic engine. The largest decline, in the personal services area, reflects less willingness to spend on things that could be done by the individual. However, this doesn’t extend to areas where the service is not directly paid by the consumer. Healthcare, where job counts are increasing explosively, is largely covered by insurance benefits or government programs. Online shopping impacts office jobs by taking retail out of offices/stores. In other words, automation may not be fully responsible for things we’re seeing.
Now, let’s now look at jobs in more detail. It seems logical that a place to apply new automation strategies like AI would be the specific places where jobs are increasing based on traditional approaches.
Among office jobs, management occupations, business and financial operations, and computer/mathematical occupations all show very high CAGRs relative to jobs overall; more than double the pace. These are all knowledge worker categories, and so are all jobs that could be enhanced by AI information. They are also all jobs that have a high unit value of labor. The troubling fact here is that they should have been jobs that traditional automation could have targeted. Why didn’t they? I would argue, based on casual enterprise comments, that automation in any form tends to support reducing lower-level jobs but creates higher-level jobs. More but smaller teams, more managers, remember? Automation specializes, in short, and shifts activity from human work to human supervision. This shift has reduced job count, but increased unit value of labor.
In the labor categories, we see the highest CAGRs in jobs with relatively lower unit values of labor. This likely means that automation practices to date have pulled higher-value jobs out of the market, which means that what remains is likely to be harder to target. Axios is right in saying that it appears automation has already impacted office labor, making the opportunity for productivity enhancement greater in labor, and thus the AI opportunity. However, there’s a lot of stuff that has to be done to make real-time automation capable of enhancing productivity. Is there any hope of pulling more out of the office space?
I think that the real reason why labor is a better target is more complicated than the Axios chart suggests. To get to labor productivity, we have to integrate computer technology into the work itself, not rebuild jobs around the computer. The latter process has made knowledge worker elites premium requirements, and empowering them demonstrably does not reduce their job counts. If we could use AI, and in particular AI agents, in the way enterprises want to, which is to build them into workflows, we could target the flows of human work as much as computer work, and create something that has a bigger net impact on labor cost.
Does this mean that AI replaces us? That, unfortunately, gets us back to the point about price, demand, and opportunity. The industrial revolution replaced a lot of craft workers, but it lowered goods prices enough to create more demand and the overall effects on employment and quality of life were positive. How many workers could AI displace? Could it create new spending to create new jobs of a different type? Would these jobs be accessible to the displaced workers? I don’t think this is going to be an immediate issue for us to consider, but it probably will be down the line.
