Will AI make knowledge workers obsolete? Perhaps the answer is neither a “yes” or a “no”.

In hindsight, I could have graduated art school early. Instead, I signed up for a volley of interdisciplinary classes. Among them were “Web Design 101” and “Programming for Artists”, which marked my future and the way I saw the world.

On the first day of Web Design 101, I sat with business, art and english majors as we were assailed with the technicalities of how HTML, Javascript, and CSS were responsible for the webpages we used every day. In the beams of the noon sun coming through windows of what was once a train depot, this new world of code was exciting, but in a vague, theoretical way. So I was caught off guard when our professor, smiling, ended class with the following assignment: “Submit your first website tonight, and I’ll see you Thursday.”

That night, I stared down a blank text editor (Atom!) in a cold sweat. Now, understanding a few DOM manipulations via MDN documentation feels trivial. Yet that night, it felt utterly impossible.

It might as well have been magic.

image

Writing a javascript game with Russell Boswell at the recurse center, 2025

In a way, that’s exactly what it was, and I think that very magic is the defining element of industries that are currently being redefined.

Fast forward a decade, and the arrival of commercially available, transformer-based LLMs brings with it a debate; are the human beings in this industry replaceable? I hear this question all the time. Someone literally asked it yesterday morning.

Each industry contains a contingent of professionals who embrace the changes AI offers their craft and their future. And in each industry, a rival faction quietly faces them down, armed with conviction (or hope) that the changes they see are overblown hype, or perhaps a total mirage.

image

A diplomatic point by Nicholas Bergson-Shilcock, co-founder of the Recurse Center in Brooklyn, about the existence and reception of AI skeptics as he attempts to play arbiter to these two factions

I’m simplifying. For one, each contingent is subdivide by at least one more dimension, the dimension of willingness. I visualize it roughly as a punnet square:

Excited about AI promisesBummed about AI promises
Belief in AI promisesWilling BelieverUnwilling Believer
Disbelief of AI promisesAgnostic LudditeResolute Luddite
For better or worse, when I read scan the ubiquitous editorializations of AI-fueled economic transformation, or arguments for the lack thereof, I sort them roughly into the four categories above. I do this partially to understand the motives behind each respective belief - sincere or ulterior.

I imagine the real sweaty human being (I’m not talking about you Nick, I promise) behind a circular LinkedIn mugshot posting diluted versions of their opinion only to log off, lose their smile, and drop their head in their hands.

I imagine Larry Ellison snorting a line of coke off of the back of his smart-toothbrush before putting on the flannel jammies he reserves for his northern hemisphere yacht, and thinking “Wow, AI is the best. I wonder if I can vibe code a simulation of me eating the heart of Montezuma”. He’s got sort of conquistador energy, no?

Meanwhile, there are developers like me, tinkering around with new tools, spending one month going back to writing code by hand, another leaning heavily into a Claude subscription. Noticing things have changed, and wondering how thoroughly the economy of knowledge work has.

Meanwhile on the luddite side, there are many who say this could be the worst thing to ever happen to humanity and yet in some elaborate conceit, they are the ones working the hardest at the AI labs. They seem to be reading off the same cosmically funny, pathological script OpenAI did when they named their company around their initial vision to “save humanity” only to become the epitome of unstoppable computational doom.

These different positions make for a fascinating, silly, uncertain and troubling world. I’m not here to weigh in on a debate.

I’m fascinated by how long these questions have been hanging over us.

The continued lack of resolution indicates to me that there is no path set in stone. That history and economics and human agency could proceed in many different directions.

It’s fitting that in an age defined by the supposed replacement of human thought, the value of a single human’s agency and belief feels as needed and scarce as ever.

So my question is not whether AI will take [our] jobs. My question is, what makes some jobs so much more vulnerable than others in this time?

image

Demo of a helmet-mounted lidar scanner at Side Project Saturday in Downtown Brooklyn

With am irnic smile, my friend Zach recently quipped “AI is coming the hardest for the nerds,” invoking a sort of Frankenstein’s monster situation.

Another friend, Akhil, described a commute he’d had on his way to work where he labored to make sure that chemical weapons weren’t available via casual chatbots. He’d had the pleasure of sharing a subway car with an older man enjoying the music on his JBL speaker and the purple drink in his styrofoam cup. He was bragging about all the kids and grandkids he’d had with different women while the passenger next to him tried her best to read her book.

“Now that guy,” my friend said, “Isn’t worried about AI.”

image

Designing a prototype for an interactive keyboard glove on a breadboard in Python

In that same conversation, someone else at the writing group, also a software engineer, said that he was interested in “marrying a nurse”, because it seemed like the vocation that would evade the clutches of automation the longest.

And in fact several people I know have left the engineering field to become a health worker, either because it seemed to have more professional stability, or simply because they’d prefer to have an observable (positive) impact on the wellbeing of human beings.

I could go on, but I think that in software engineering communities it is clear that LLMs are remarkably good at writing code, and unlike say, the entertainment business, the argument that “they can’t do what we do” just doesn’t feel quite believable anymore. The activity that everyone thinks of when they think of software engineering, writing code, truly can be done incredibly well by Claude or Codex at this point. The conversation around job replacement in tech has turned more to understanding the less tangible activities and qualities engineers have that AI is still not great at, of which there are a still surprising number.

But why software engineering? Why is AI coming for the nerds and not the nurses?

My answer is that human beings have always had magic and always will, but what that “magic” currently is is always changing, like a societal symlink. For a while, it was software. How does the internet work? How do you create distributed systems? This is of course not my idea, but Arthur C. Clarke’s famous “third law” for for sci-fi authors:

“Any sufficiently advanced technology is indistinguishable from magic.”

image

Earscratch break from shipping healthtech

I think there may be a caveat though, which is that magic stops being magic when ChatGPT can write the spells on the free plan. When part of it is understood, the magic mischievously moves, deeper into the obscure reaches of the collective understanding of humanity. And when a field is totally understood, it disappears entirely. It’s sort of like Syndrome’s line from the incredibles:

Error uploading file: Request failed, status 400

Syndrome, ironically, a tech whiz though. Also, I have to say, his evil plan was actually kind of democratic. It’s hard to be super mad when you think about it too much.

It’s the same endless treadmill globalization has always been so good at advancing. Value is identified, understand, industrialized, marked down, and then taken for granted.

Of course, this line of thinking breaks down in some important ways. LEDs are ubiquitous, nearly immortal, and dirt cheap, and yet I probably could tell you more about how to manufacture a combustion engine than a set of ferry lights.

But code isn’t quite the same. It’s built on text editors, and generally compiled or parsed with freeware. You don’t need special minerals. All you need is special knowledge, and a pipeline has now been built that mainlines that knowledge to anyone who wants it.

It’s sort of fascinating to imagine the hurtles that still exist for the production of the software engineers. Time, volume of knowledge, experience of failure modes. Confidence in the esoteric, which is maybe not so different from interest in the esoteric. Some say “taste”.

And so I think there is a fundamental question for knowledge workers and non-knowledge workers, and it is “do you want to be a wizard”? Do you want to base your livelihood on popularly celebrated obscurities? If so, then the magic to chase is probably machine learning and the mathematics that props of AI research and safety.

If you are content with less powerful obscurities, then you are free to seek out new genuses of previously undiscovered parasitic wasps, becoming more of a druid - less of a Saruman, or a Gandalf, and more of a Radagast. No tech companies that manufacture drones will ever be named after Radagast.

image

Radagast, the druidic wizard in Tolkien’s Lord of the Rings, who is powerful in his own right but is content to tinker with animals and trees in the Fangorn Forest.

And if obscurity and magic are entirely without appeal to you, that’s very okay. Become a nurse. Become a plumber. Become a carpenter. An electrical engineer. Become a journalist, if you don’t care about money at all. Or a cook, or sell shoes. There will always be magic, and we will always worship it, but that doesn’t automatically mean it’s important, or that there is any more integrity to it.

In conclusion, I think that roles in knowledge worker industries like software will continue to change and evolve, but not disappear. The ones that will change the most are the ones that are the most “magical”, because of the roles in which obscurity provided their value. While that obscurity is made less valuable by new technologies like LLMs, what will be left over is the harder work that isn’t sexy or mysterious.

I think that some of the most interesting disruptions caused by AI are yet to come in the form of enormous leaps and bounds in various industries, but these events will generally punctuate long plateaus of relative stagnation.

Take healthcare as an example. This industry needs an enormous amount of work to be done on it all around. Engineers continue to be needed. This isn’t “disruptive”, or state of the art work.

In other words, it’s not magic. It’s just hard work. And so the magic will pass over it, moving on as it always days. But in its wake, crucial work will remain in every industry.