AI Is Coming For White-Collar Workers
Assume that artificial intelligence (AI) will take your job. Precautionary actions to upskill must be taken now to avoid replacement. This is true unless you are a physical laborer or in services that rely on physical human-to-human interaction.1
Most people understand on an instinctual level why AI could be a threat to their job. The fact that AI is getting better and better at graduate-level mathematic problems, real-world coding challenges and real-world merge requests, financial planning, so-called "deep research", and medical diagnoses does not need an explanation to connect its relevance to the replacement of common human work tasks.
People know the aforementioned tasks are key to their work, that AI is quickly becoming superior to some of the best humans, and that if an AI could be well integrated now, then they could become less valauble or even unvalued. But the conventional widsom, held by many of the 70% of software engineers who do not view AI as a threat to their jobs (StackOverflow, 2024 Developer Survey), suggests that AI is merely a productivity enhancer, and for myriad other reasons it poses no real harm.
I enumerate some of the more common reasons below:
- "AI isn't creative like humans are."
- "AI might be great, but a human ultimately has to verify the results."
- "AI is already so good, and it hasn't taken my job yet, so it's pretty unlikely to in the future."
- "AI is a catalyst for human production, and the demand for my job is near infinite, and so long as I can use AI to produce more for my company, my company will want more, and I'll still be valuable."
These are all bad reasons for a fearlessness of AI.
1. Humans Are Not as Creative as You Think
Whenever someone suggests that a human is creative, they typically evoke the image of a pioneering artist's magnum opus. They call to Edvard Munch's The Scream, Homer's The Odyssey, or Orson Welles Citizen Kane. This fails to account for the fact that each of these artists are mere imitative iterations on, adding only slight permutations to, artistic lineages spawned hundreds of years before them. Each artist exercises inspiration largely drawn from their personal lives (one form of data) and other art they admire (another form of data). Munch was inspired by Vincent van Gogh and Paul Ganguin; van Gogh was inspired by Millet, and Ganguin was inspired by Monet. So on recursively go generations of training data.
Notice that when humans attempt to depict aliens, be it the Klingons of Star Trek or the Heptapods in Arrival, alien subjects with no available training data, they are almost always anthropomorphized or bestial; humans hallucinate mediocre guesses based only on what they know: One alien looks like an orange human with ridges on their face; the other alien looks like an octopus with claws.
More than likely, you did not invent drawing, multiplication tables, poetry, love letters, kisses, furniture, or really much of anything at all. Like artists, you are the product of human iteration over millenia. Who is to say that AI can not iterate in much the same way over the same number of years? AI has just gotten started. If humans can be deemed creative, it stands that AI could one day be considered creative as well.
2. Verification Is Not Innately Human
A common argument is that humans need to verify the product that AI produces. This is not so. All humans are fallible, and all humans have a non-zero failure rate, even when they are utterly competent.
You can take one of the most gifted software engineers in the world, and they might have a failure rate of 0.001% when it comes to approving code as "bug-free" and suitable for safety-critical missions, but as soon as AI can devise tests or use reasoning to approve bug-free code with a (0.001% * 0.9) failure rate, that gifted engineer will be replaced. Companies are agnostic as to whether a human or an AI does a job; they care about the product.
With white-collar jobs, the problems being solved are also usually more verifiably correct or incorrect. A mathematical expression can be proved to have a certain solution. Code can be proven to work as intended. A bridge can be shown to withstand certain winds. Once AI's work can be formally verified in a given domain, even a perfect human verifier, not just the engineer, risks obselescence.
3. Change Takes Time but Occurs Nonetheless
Industrial automation and international competition did not corrode the United States's Steel Belt overnight. The change from Steel to Rust was gradual and pernicious. If you asked a manufacturing employee in 1950 whether they felt that international headwinds or automation would have cost them their job, they likely would have balked at the idea. In the year 1950, 43% of jobs in the United States were in manufacturing and over 20% of all jobs were in manufacturing in the Rust Belt; in the thiry years after 1950, 34% of the Rust Belt's manufacturing jobs disappeared (Minneapolis Fed). And as of January 2025, less than 9% of total jobs in the United States are in manufacturing (Univ. Wisconsin, SP).
AI has a high chance of making even faster and greater change.
4. The Demand for Your Job Is Not Infinite
The demand for your job is finite. There is not an infinite demand for any job because no one job can produce a single resource that can satisfy all needs. If all the world was comprised of a single profession, humanity would starve to death or die of thirst. Due to constraints, demand is distrbuted across many fields at equilibria points. Humanity needs a certain proportion of carpenters, electricians, farmers, doctors, software engineers, and so on to thrive. These proportions are susceptible to change, but only so much. To think that one job in particular does not obey this law, and companies will try to infinitely increase supply through said job even when there is not enough demand to meet it, is incorrect.
Companies will do what they can to cut costs and increase productivity within their limited budget and for their limited market demand. Soon hospitals, law firms, software companies, consultancy firms, schools, and all manner of industries will be substituting their names into the following description of the Rust Belt's demise (Brookings):
In successive waves of industrial restructuring, employers shuttered inefficient factories, moved production to cheaper locales, automated where possible, and pushed costs and risks onto employees and suppliers.
Upskill Now
AI requires integration and infrastructure, which must be engineered over years, and it must be adopted, which likewise takes years, but the fundamentals of economics guarantee us that AI is will likely be primed to take your or your friend's job. Businesses that do not adopt AI will lose to companies that do, much like companies that did not adapt to the internet revolution either lost their dominace or perished. And if you do not heed this warning and your job is spared by AI despite not upskilling, it might just be dumb luck. Upskill while and if you still can.