There’s a growing attitude in the technology industry that LLM technology is, or will be, the next great innovation to our work. Business owners and workers alike seem to be in unlikely agreement: owners are thrilled at the prospect of making their businesses more efficient and productive, and workers are thrilled at the prospect of simplifying their workstreams or making their jobs easier. It’s a surprising alliance in the prevailing economic system.

This is a post about the economic implications of adopting AI in your work, not about whether or not they actually improve your productivity. For the sake of argument, this will assume there is some productivity added. There is plenty of criticism of countless aspects of AI/LLMs/et al technologies online, but this post engages primarily with the concerns of labor.

Some context on capitalism

From Andrew S. Zimbalist’s Comparing economic systems: a political-economic approach:

Capitalism is an economic system based on the private ownership of the means of production and their use for the purpose of obtaining profit.

That obtained profit can come from many places: production facilities expansion, resource availability, or technological advances make production cheaper, macroeconomic forces increase demand for a product, allowing its price to be increased, etc.

However, in the recent past, that profit has been disproportionately extracted from the labor force. The Economic Policy Institute analyzed productivity and wage data from Bureau of Labor Statistics and Bureau of Economic Analysis from the 1940s until early 2010s, and found that while productivity increased hand-in-hand with hourly labor compensation from 1948 to 1973, the next 40 years showed an increasing and immense gap: while productivity went up nearly 75%, hourly compensation only increased 10%.

This can be attributed to the mechanics of the system. In a capitalist framework, businesses must create profit to be considered successful, and extracting more and more of that profit from labor is practically codified in the economic playbook of capital: stagnating wages, off-shoring work to laborers with lower standards of living, even leveraging legal systems to allow for fewer labor affordances. Capital will always choose the interests of capital over the interests of labor — it must do this to survive.

Where LLMs come in

Capitalism as it stands, in America and beyond, only functions if it continually finds new ways to extract value from labor: so while the upsides of adopting LLM technology to you as a worker might seem marginally useful now, that adoption directly contributes to that extraction, ultimately devaluing labor.

If we take the lofty claims of LLM marketers to be true, that by adopting LLM workflows, you may do the work of your team, say 25% faster, what are the implications of this to the managerial/supervisory class? Consider a scenario:

Let’s say your team is expected to create 1 widget per period, and your team has 5 people on it. By the time your entire team is 25% more efficient, you have gone from individually creating .2 widgets to creating .25 widgets with the help of LLMs, which means your team creates 1.25 widgets per period.

Applying the systematic behaviors of capitalism, one may envision two possible outcomes from this:

• The first likelihood is that you now are paid the same wages and expected to create more widgets, which means your labor has necessarily gotten less valuable. If you got hired at $100 per period to create .2 widgets, and now get paid $100 to create .25, you functionally get paid less ($20 less per period in this case, but it’s academic).
• The second likelihood, and the more insidious one, is that the increase in efficiency justifies downsizing. If your team is expected to create 1 widget, and now creates 1.25, the bottom line to the company is that if it lets you or one of your teammates go, it can create 1 widget for 25% cheaper. That’s free profit to the company, and sacrificed employment to one of you.

While to an individual that loss of employment might be devastating, to the company a reduction in workforce is a line-item; as mentioned before, capitalism will always choose its own interests. It’s not a personal decision, it’s not a specific person that can save you or betray you: capitalism, as a system, was designed to work this way.

There is a farther-reaching consequence to this, which is that: the more roles that companies eliminate under this increase in efficiency, the higher the supply of the labor market, which as market economics 101 will tell you means that the price for labor will go down, worsening the crisis of stagnate wages.

What is there to do?

As technologists, we are workers first. We cannot individually stop the machine of capitalism from exploiting our labor, and, while this tide will shift, many of us are too romanced by the technofuturist dream of LLMs — despite widespread, multidisciplinary criticism of the technology and disastrous consequences in its application so far — to easily reconsider our adoption of it.

The solution to the oncoming waves of LLM-related labor devaluation has at least two solutions, which should be applied in tandem:

As individuals, we must reject the indiscriminate proliferation of LLM technology in our work. As many have written, there may actually be actually valid applications of this technology, but any that weaken your role as a laborer should be viewed with intense scrutiny. The immediate convenience of writing fewer TypeScript interfaces will be followed swiftly by the economic ramifications of cheapened labor — no matter how much you like your boss, it is a systematic certainty in a capitalist market.

Collectively, we must organize as a labor force. Countless examples through the history of capital show that production owners/business people have no qualms about organizing against us — from the Pullman Strikes, to the Ludlow Massacre, to the Mohawk Valley Formula, the class of means has no incentive not to collude, organize, and act against the interests of labor. Labor must consequently organize and act in its defense.

There is great precedent and inertia for labor organization in the technology industry — I recommend Ethan Marcotte’s book You Deserve a Tech Union as further reading.

Thanks for taking the time :) I’d really like this to be a strong if succinct advocation, so feel free to email me if you have any questions or concerns.