I have two friends. One works as a graphic designer, the other as a data scientist.
The designer chose design because of taste. She loves that feeling of turning a vague emotion into a concrete image, loves studying how colors combine, how light falls, how composition makes an ad catch the eye in two seconds. She chose this profession because those things excited her. What does her daily work look like now? Her boss hands her a brief with exact dimensions, she opens Photoshop, extracts the assets in the approved style, arranges them, and exports three sizes of cuts. After half a year of this, she said she felt like a slow Photoshop script.
The data scientist friend had the same trajectory. She chose this path because she believed data could directly shape business judgment, and that mattered. Once on the job, 80% of her time went into writing SQL, and the rest into debugging someone else’s SQL. A year in, she was very good at writing SQL. But the thing she originally wanted to do, understanding the business through data and helping the company make better judgments, was not happening any more than on her first day. She had become a faster SQL machine.
What moves me about AI is that it can take away exactly the things they never wanted to do in the first place. Let AI write the SQL, and she focuses on what the data reveals about the business. Let AI handle the Photoshop cuts and layouts, and she focuses on what emotion users carry into this context and how a design can reach them. What AI does is bring them back to the part of the work that excited them in the first place.
This is not just a story about AI making things faster. It is a story about how industrial division of labor distorted the shape of most professions, and how AI can undo some of that distortion.
Let me be clear about one thing. I am not saying mechanical labor has no value, and I am not saying all mechanical work should be eliminated. The daily substance of any profession consists of two kinds of work. One is mechanical: following rules and procedures, with little discretion over outcomes, where skill mostly means speed. The other is judgment: understanding context, making trade-offs, drawing conclusions under incomplete information, where the quality of the outcome depends heavily on what happens here. Every profession is a mix of the two. What varies is the ratio.
A staff-level engineer writing code spends most of her time understanding the business, designing interfaces, weighing trade-offs. The mechanical portion of actually typing code is small. A surgical resident on call spends most of her time suturing, tying knots, writing notes. The judgment portion is relatively small now, but the weight of judgment grows rapidly with seniority in this profession. An assembly-line worker’s job is almost entirely mechanical. These are three very different jobs sitting at different points on the spectrum.
What lets people enjoy their profession, grow in their profession, and create real value, comes almost entirely from the judgment end. The mechanical end sometimes carries judgment as a vehicle (I’ll return to this), and sometimes it is pure cost. Either way, what makes a job its own particular thing is the judgment part.
The problem is that industrial division of labor has been pushing most professions toward the mechanical end for about a century. The force behind this push is that mechanical work is easy to manage. Judgment is hard to define as KPIs; mechanical work is easy. Judgment is hard to hire for; mechanical work is easy. Judgment is hard to verify in output; mechanical output you can just count. Companies, as an organizational form, have strong incentives to break work into standardized tasks and turn every role into a link in a production line. This breakdown is good for the company’s management cost and disastrous for the daily experience of the person doing the work. A job that used to be 50% judgment and 50% mechanical becomes 80% mechanical and 20% judgment after the breakdown. The state of my two friends is the result.
What AI does here is make it possible to dial the ratio back. If a data scientist used to spend 80% of her time writing SQL, AI can now compress that down to 20%, which means the other 60% can go back into judgment work. This is different from “AI makes things faster.” Faster means more SQL in the same time. Rebalancing means the same time goes back into judgment. For the person doing the work, the felt difference is enormous.
The argument above applies straightforwardly to people who are already in the profession: their work was distorted by industrial division of labor, and AI undoes some of the distortion. But what does the same mechanism mean for new graduates and junior practitioners? This needs a separate treatment, because there is a popular objection around this point.
The objection goes roughly like this: AI has taken over the low-level mechanical work, so new graduates and junior practitioners lose their chance to grow, because those seemingly mechanical jobs are the necessary path to accumulating experience. Write enough SQL and you’ll learn which data is dirty, which queries are dangerous, which metrics fall apart under scrutiny. Work enough with Photoshop and you’ll know the engineering constraints of color in different contexts, the differences between print and screen, the edge cases that break at odd resolutions. These are the foundations of senior judgment, and if AI takes them away, juniors lose the path to senior.
This objection sounds reasonable on the surface and connects to a real social problem, the career ladder for young people. But if you look closely at how junior practitioners actually grow, its premise does not hold.
Here is what I’ve observed. If a junior data scientist spends eight hours a day writing SQL, what actually makes her grow is the last one or two hours where she stares at a number and says “this doesn’t look right.” It is not the six hours of typing keys, looking up syntax, and debugging bracket matching. Why is this DAU 30% higher than last week? Why is the conversion rate on iOS twice that on Android? Why does this cohort’s retention curve have a strange drop on day seven? These questions are where judgment comes from. And whether she hand-wrote the SQL or not has nothing to do with asking them. As long as she can read the query and understand what it does, trace where the data came from, and dig back into the source when the results look off, she can build that judgment.
Put differently, accumulating judgment and executing work by hand are separable things. What she needs is exposure to real business messiness: dirty data, mismatched definitions, counterintuitive results, awkward business concepts. Those are the real vehicles for judgment. Hand-typed SQL is just a byproduct of industrial division of labor over the past few decades.
This view can be checked against a few analogies. A pilot’s judgment comes from repeatedly making decisions in real flight situations, not from building the plane or refueling it. A head chef’s judgment comes from repeatedly handling fire, taste, and timing under real service pressure, not from plucking chickens and washing vegetables. A lawyer’s judgment comes from repeatedly analyzing legal relationships, assessing risk, and designing strategy in real cases, not from printing contracts and binding files. All these professions have mechanical work at the bottom, but that mechanical work contributes very little to the growth of judgment. A junior pilot needs enough flight hours and enough variety of weather and scenarios, which has little to do with how skilled her walkaround inspection is.
Data scientists and designers are in similar territory. SQL and Photoshop are tool-level physical labor whose value lies in carrying the context of judgment. As long as junior practitioners can understand what these tools are solving and trace back when something goes wrong, the path to accumulating judgment stays open.
Going further: under industrial division of labor, junior positions dilute a small amount of genuinely valuable judgment training with a huge amount of meaningless mechanical work. A junior data scientist’s eight-hour day contains only one or two hours of actual judgment training. The other six or seven go into handwork that contributes almost nothing to her growth. If AI can take away those six or seven hours, in principle she could spend all eight on judgment training. A junior with AI as the execution layer could grow several times faster than a junior with a human execution layer, provided she actually spends the reclaimed time on questioning results, understanding business, and wrestling with data, rather than scrolling short videos.
So the claim that “AI blocks the career ladder for new graduates” has the causality backward. AI is eliminating positions where the work has little to do with judgment. What disappears was always positions that had no independent value for the growth of junior practitioners. Companies kept those positions because labor was cheap, not because they were good for the juniors. That was just one way of using cheap labor.
Looking at both cases together, they are two sides of the same thing. For those already in the profession, AI returns the judgment time that industrial division of labor squeezed out, letting them do the part of the work that drew them to this profession in the first place. For junior practitioners, AI lets them start accumulating judgment directly, skipping the years that earlier generations spent being alienated by tools. The first is returning to the source; the second is not having to take the detour. The mechanism is the same; only the life-cycle position differs.
Does this hold for every profession? No. There is a class of professions where the low-level mechanical work itself carries the entire path to accumulating judgment. Take that mechanical work away, and there is no way to grow into judgment.
The surgeon is the clearest example. A surgical resident suturing wounds, tying knots, making basic incisions and closures, is doing something that looks mechanical but carries a sensitivity to tissue firmness, a sense for where blood vessels run, an adaptation to individual patient variation. This feel cannot be learned by watching. It has to be accumulated by your own hands making the same motion thousands of times. AI can assist diagnosis, help read images, help plan surgeries. But the tactile feel AI cannot take over, because the learning channel is in the hands themselves. Woodworking is similar, so is cooking, so is piano.
The rule is simple: if the value of low-level mechanical work comes mainly from tactile or embodied experience, AI cannot take it over. If the value comes mainly from the cognitive purpose it serves, and that purpose can be reached through other paths, AI can. Designers, data scientists, programmers, lawyers, and most white-collar work fall into the second category. Surgeons, carpenters, and chefs fall into the first. AI’s impact on white-collar work is larger precisely because white-collar alienation runs deeper, leaving more distortion that can be corrected.
I started with my two friends, one a designer and one a data scientist. They chose those professions because of the parts that excited them. After they entered the workforce, their daily lives became being used as machines. This is what industrial division of labor has done to almost every white-collar profession; it is not their individual problem. What AI does is pull that distortion back a little, letting their work look more like what made them choose it in the first place.
For people already in a profession, AI helps them find again the reason they chose this work. For people entering a profession, AI lets them start directly from judgment, skipping the years of being alienated by tools. Both are a better state for the people involved.