Cover essay
On the embarrassment of expertise
AI did not replace the professions. It separated what they were paid to do from what they were paid to know.
There is a feeling that has not had its name yet, and we suspect it will be the central feeling of professional life for the next decade.
It begins, as these things tend to, in a corner office on the forty-second floor. The head partner — three decades at the firm, walls lined with framed announcements of every deal he had ever closed — calls his newest hire into his office. He has something for her. Their largest client needs a draft contract for an acquisition. It is the kind of work that, in his memory, defined whether a young lawyer survived at a place like this. Late nights. Coffee gone cold. The slow accumulation of clauses, exhibits, and the kind of fine print that protects a hundred-million-dollar deal from a single misplaced word. He gives her until Friday. He half-expects to see her again Wednesday, hollow-eyed, asking a question that at least proves she has been reading.
She comes back in an hour.
He does not look up at first. He assumes she has a clarifying question, or worse, that she has frozen — that the blank page has done what the blank page always does. When she sets the document on his desk, he glances at the page count and feels the first irritation. Sixty-two pages. Fully formatted. He thinks: she has copied something, or she has missed the point entirely, and either way he is about to have an unpleasant conversation with the recruiting committee about the quality of this year’s class.
He starts reading. He intends to mark it up in red — the kind of bleeding pages that teach a young lawyer humility. By page four, his pen has not moved. By page nine, he has forgotten he was holding a pen. The promises each side makes to the other are airtight. The clauses that govern what happens if the deal sours anticipate his client’s specific industry risk in a way he, frankly, would have had to think about for an afternoon. The way she has handled which obligations survive after closing — which carry, which expire, when — is, irritatingly, a little better than how he would have done it.
He reaches the end. He sets the document down. He looks at his newest hire — the one who, it occurs to him, has been at the firm for eleven days.
“How the hell,” he says, “did you do this?”
She tells him she used Spellbook. He has heard of it. He has, in fact, told three different bar association panels that programs like this were toys, that judgment was not the kind of thing one delegated to software, that the profession had a runway of years. He is a careful man, and he had been careful when he said it.
He sits with the draft on his desk for a long time after she leaves. He is not thinking about her. He is thinking about the floor below him, where eighteen senior associates are billing a thousand dollars an hour on documents that, he now suspects, an articulate twenty-four-year-old could produce before lunch. He is thinking about the client on the forty-fourth floor of a tower across the city — the one with an in-house team, with interns, with a procurement department that reads invoices very closely. He is thinking about every conference room in every office building in every city.
The feeling does not have a name yet. But the question that follows him home tonight is the question that will follow many of us, regardless of trade: what, exactly, was I being paid for?
We were not being paid by the hour. Nobody was.
That was the trick of the billable hour, and of every other hourly fee on every other invoice — it made it look as though hours were what we sold. They were not. Hours were the unit; expertise was the product. The hour was a proxy, and like every proxy, it worked until something arrived that broke the relationship between the proxy and the thing.
The newest hire, in our story, broke the proxy. She produced sixty-two pages in an hour, and the pages were clean. By the standard of whether the labor happened — yes, it happened, and it happened roughly forty hours faster than any associate of his could have produced two years ago.
But notice what the partner did when he sat with the draft. He did not re-read it to admire her phrasing. He sat there because he was thinking about three things she did not, and probably could not, know: the way this client had reacted to a similar clause in a deal six years ago. The fact that the regulator who would review this transaction had been writing irritable letters to firms in this industry for the last eighteen months. And the unspoken understanding, never written into any retainer, that this client’s CEO did not want surprises, ever, on page forty-seven.
That is what the partner was being paid for. He was not being paid to type. He was never being paid to type. He was being paid to know which of the sixty-two pages mattered, in this room, for this client, this quarter, and to be able to pick up the phone and explain it. The labor was the wrapper. The expertise was the contents.
This is the distinction that AI is now making impossible to hide.
For most of the history of professional work, the labor and the expertise were welded together. We could not have one without the other. Producing a clean draft contract required knowing enough to produce a clean draft contract. The hours were not a fiction — they were the only available delivery vehicle. Wrapper and contents shipped in the same box because they had to.
That is no longer true. The wrapper now ships separately. Spellbook, and the dozen tools like it, and the dozens more coming, can produce the wrapper. Anyone with a thirty-dollar subscription can produce the wrapper. The wrapper, increasingly, is free.
The contents are not. If anything, they are worth more, because they are now the only thing left on the invoice that can justify the invoice.
Here is the test. Hand a thirty-dollar subscription to a smart, motivated person outside the field. They will produce a document that looks, to another person outside the field, indistinguishable from the work. This is genuinely true and worth admitting. It is also irrelevant. They will not know which clause is the one that matters. They will not know what the regulator has been writing letters about. They will not know what page forty-seven needs to look like for this CEO to sleep tonight. They cannot ask the tool the right question, because they do not know which question is right. They cannot audit the answer, because they do not know what a wrong answer looks like.
That is the entire game now. Expertise is no longer the ability to produce the work. Expertise is the ability to direct the work and audit it — to know what to ask for, and to recognize, in a stack of plausible output, the one detail that would have ended a career in 2015 and may still end a deal in 2026.
This rearranges who is valuable and how. The partner is, at the end of his long afternoon, more valuable than he was the day before, not less. His judgment has not depreciated; his typing speed has. The newest hire is in a different position. She can produce partner-level wrappers, but the wrapper is not what she is paid for — and the contents she is paid for, she does not yet have.
The question we were left with at the end of the last section was what, exactly, was I being paid for? Here is the answer. We were being paid for what we know, and for the part of what we know that we can articulate clearly enough to direct an instrument that does not get tired. When we can articulate that, the instrument multiplies us. When we cannot, the instrument exposes us. The hour was always a fiction. The fiction is just over.
It is tempting, having looked at the partner, to file the unbundling under “professional fee compression” and move on. We are not going to do that, because the same machinery is now visible in places where the stakes are higher than money.
Money is the easy case. When the wrapper detaches from the contents, fees compress, professional pyramids re-shape, and a generation of mid-career people negotiate the next chapter. Painful, but ordinary. The harder cases are the ones where the wrapper has been hiding, all along, the part of the work that keeps people alive, or keeps a society running. We want to look at two of those.
The first is medicine, and we will use radiology as the example, because radiology is where AI’s capability has been most visibly real for the longest. The newer detection tools are genuinely good at finding things in imaging. They can read a chest film, a mammogram, or a CT scan and flag what looks abnormal in seconds, with accuracy that, on a wide range of conditions, now matches or exceeds the average human reader. The labor compresses. The wrapper detaches. So far, the story rhymes with the partner.
Then it diverges, because of something that does not happen in a contract.
What is starting to be documented in radiology, and what we believe is the early signal of a much larger problem, is deskilling. When a radiologist works alongside an AI tool that pre-flags findings, her own pattern-recognition — the slow, unconscious accretion of thousands of films read in residency and beyond — begins to atrophy. The skill that used to be exercised every shift gets exercised only when the AI is uncertain or absent. A growing literature suggests the effect is real and measurable.[1] The radiologist who has been in the loop with AI for two years has become, in some studies, a worse reader of the cases the AI handles confidently — and those are not the cases that matter. The cases that matter are the ones the AI gets wrong: rare presentations, atypical anatomy, the patient whose imaging looks like one thing and is, in fact, another.
This is the lethal version of what the previous section described. There, we said expertise is the ability to direct and audit. In radiology, the expertise required to audit is the same expertise that erodes when you stop being the one who looks first. The wrapper and the contents were not welded together by accident — the daily exercise of producing the wrapper was what kept the contents sharp. A field that hands its labor to AI without asking how its experts will stay expert is a field that is, quietly, losing the ability to catch the cases AI cannot.
This is not an argument against the tools. The tools are good and patients benefit from them. It is an argument for redesigning the work, the training, and the audit trail around a different premise. The premise of the last fifty years was that doing the work was how you got good at the work. That premise is gone. Whatever replaces it has to do, deliberately, what daily labor used to do by accident.
The second case is software engineering, and we are aware that this sounds, on its face, like a smaller problem than radiology. It is not. It sits one layer beneath the others.
Software is the substrate. The hospital scheduling system that decides whether the radiologist sees a film at 7:14 a.m. or 2:40 p.m. is software. The pricing engine the partner’s client uses to value the target is software. The traffic light at the intersection on the corner is, increasingly, software. When AI tools begin producing software at the speed and scale they now produce it, the labor of writing code compresses dramatically — and the question of which code gets written, and whether it should have been, becomes the entire job.
This is what we mean by the agency case. The engineer is no longer being paid to write the function. Anyone with a thirty-dollar subscription, including the engineer’s own tooling, can write the function. The engineer is being paid to decide what the function should be, what it must never do, and what happens to the world when it runs. That has always been the actual work. It used to be hidden inside the labor of writing code, the way the partner’s judgment was hidden inside drafting the contract. The hiding is over.
When the wrapper compresses in software, decisions get made faster, by people closer to the keyboard, on shorter loops. The senior engineer who used to be the friction in the system — the one who said wait, what does this break? — becomes a delay to be routed around. The mistake that used to take a week to ship now ships in an afternoon. And because software is the substrate, the mistake does not stay in software. It surfaces in the hospital, in the deal room, at the intersection.
The partner’s story, at the start of this piece, was never really about a lawyer. It was about a system in which the visible work and the invisible work had been welded together, and an instrument arrived that could perform the visible work alone, and the invisible work — the judgment, the audit, the pattern-recognition, the who-decides — was suddenly exposed for what it had always been: the only thing on the invoice that mattered. That story is now being told, simultaneously, in a reading room, in a code review, in a procurement meeting, and at the intersection on the corner. The fee compression is real. It is also the least important part.
By now the question is no longer whether the unbundling is happening. The question is what people are doing about it, and what the cost will be when it turns out that none of what they are doing is enough.
The first response is refusal. Through 2023 and into 2024, the dominant institutional posture across regulated industries was containment. JPMorgan banned employee use of ChatGPT in February 2023, citing compliance risk; several other large financial institutions followed in the months that followed. In May 2023, Samsung joined the list after an internal incident in which engineers had reportedly pasted proprietary source code into the tool — Samsung’s threshold was higher; it took a leak to push them across it.[2] These bans were not, on their face, a denial that AI mattered. They were, on their face, a sober response to a tool whose security boundaries no enterprise IT department had yet figured out how to draw. But functionally, they served as a reason not to think hard about the deeper question, which is what the tool, properly bounded, was about to do to the work itself. Many of these institutions have now built internal versions — JPMorgan’s LLM Suite, expanded to roughly 140,000 employees by September 2024, is the most visible exemplar — but the gap between banning a chatbot and redesigning a profession is a chasm that none of these firms has yet seriously crossed. Refusal bought time. Time was used poorly.
The second response is substitution: the assumption that, since AI compresses labor, an organization can shed headcount by an amount proportional to the labor compression. The canonical public case is Klarna.
In February 2024, Klarna announced that its AI customer-service assistant was performing the work of approximately 700 full-time human agents, with reductions in average resolution time and an estimated forty-million-dollar profit impact.[3] CEO Sebastian Siemiatkowski described the deployment, through 2024, as a model for how digitally-native companies could re-imagine staffing. By May 2025, the company had publicly reversed course. In an interview with Bloomberg, Siemiatkowski acknowledged that customer satisfaction had suffered and that Klarna was rehiring human agents to backstop the AI; in his words, the company had let cost become “a too predominant evaluation factor.” This is the cleanest available case, in the public record, of substitution as a strategy. It went, at internet speed, from celebrated to walked back, and the walk-back was performed publicly by the same executive who had done the celebrating — which is to his credit, and which is also why we can write about it.
The lesson of the Klarna case is not that AI cannot do customer service. It can do a great deal of customer service, often well. The lesson is that the work — the wrapper and the contents, in the language of section two — was not what an org chart said it was. The thing customers were buying, when they spoke to a human agent, included an audit function (Was the AI right? Did it understand the actual problem?), an exception-handling function (What about the cases the system cannot route?), and a relationship function (Will this company show up for me when something is wrong?). When substitution removed the humans, it removed those functions invisibly — until customer-side metrics surfaced what had been lost. The headcount was reduced. The work was not. The work simply moved to the customer, who is unpaid, untrained, and free to leave.
The third response is augmentation. Augmentation is the most thoughtful of the three, and it is the response most major professional service firms have publicly adopted. Allen & Overy was among the earliest, deploying the Harvey legal-AI platform to thousands of lawyers across its global offices starting in 2023; the firm has since merged with Shearman to form A&O Shearman, and has continued to publish openly about how its lawyers are trained to use AI as a drafting and research instrument rather than a replacement. The Big Four have followed analogous playbooks. The framing is consistent across firms: AI is a tool that multiplies senior judgment; it is not a substitute for it; the partner does not become obsolete, the partner becomes more leveraged.
This framing is correct as far as it goes, and it is significantly better than the first two responses. It is also consonant with the argument we made in section two — that expertise is the ability to direct and audit, and that an instrument which does not get tired multiplies the value of expertise that knows what to ask for. Augmentation, done seriously, is the response professional firms ought to be running at.
But augmentation, as currently practiced, has a hole in it that none of its proponents has yet filled. The hole is the one we named in the radiology section: if the daily exercise of producing the wrapper was what kept the contents sharp, and if the new arrangement hands the wrapper to the instrument, then the question of how the next generation of senior judgment is formed — how today’s juniors become tomorrow’s experts — is suddenly an open one, and it is being answered, in most firms, by improvisation. The augmentation-first firm has its senior partners directing the tool more effectively. It does not yet have an answer for what its associates will know in fifteen years that today’s partners learned by doing the work the tool now does.
That is the question we will spend the next section on. None of the three responses has reached it. Refusal does not see it. Substitution forecloses it. Augmentation gestures toward it and then, mostly, hands it back to the HR department. It is the question that decides whether the professions, as recognizable institutions, exist on the other side of this transition. We think it is the most important question the essay has named.
We have arrived, finally, at the question that will decide whether the professions exist on the other side of this transition: how does the next generation of senior judgment get formed when juniors no longer do the work that used to form it?
The honest answer is that no one yet knows. We are going to spend this section laying out why — and what is at stake in the not-knowing — and we are not going to pretend, at the end of it, that we have solved the problem. The honesty is the section’s argument. Anyone who claims to have figured this out in 2026 is selling something.
Begin with what the apprenticeship system actually was. For the past century, the dominant model for producing senior judgment in the professions was a long, slow accumulation of cases. The first-year associate marked up the contract she would not be allowed to draft until her fifth year. The third-year resident read the films the attending then re-read alongside her, calling out, in real time, what she had missed. The junior engineer wrote the small function, watched the senior engineer reject it, watched the senior engineer rewrite it, and absorbed, by exposure to the rewrites, what good looked like in their codebase. The mechanism was not the formal training. The mechanism was the labor. The seniors were forged by ten thousand corrected drafts, and the drafts were the work the firm was being paid to do anyway. Apprenticeship and revenue were the same activity.
The new arrangement breaks this almost completely. The junior associate no longer drafts the contract; the instrument drafts the contract, and the junior reviews the instrument’s output. The third-year resident no longer reads the film cold; the AI flags four candidate findings and the resident ratifies, contests, or escalates. The junior engineer no longer writes the small function; she prompts a coding tool, accepts the output, and ships. In each case the junior is still working — she may, in raw hours, even be working harder — but she is doing fundamentally different work. She is auditing, not producing.
The two activities feel adjacent. They are not. Auditing requires a model of what a wrong answer looks like, and that model is built almost exclusively by producing wrong answers, having them corrected, and producing fewer wrong answers next time. The auditor without the production history is the inspector who has never welded, the editor who has never written. They can pattern-match what looks bad on the surface. They cannot reliably catch what is quietly bad — the case where everything looks correct except for the one thing that matters.
What is being tried, in firms that are taking the problem seriously, falls into roughly three buckets. The first is apprenticeship by audit: structured programs in which juniors review AI output under senior supervision, with the senior using the review session itself as the teaching moment. This is a sensible response, but it has a scaling problem and a depth problem. The scaling problem is that senior time is the bottleneck the firm was trying to alleviate by deploying the AI in the first place. The depth problem is that auditing is a thinner slice of the original apprenticeship — the junior catches some lessons by reviewing, but loses the lessons that came from the felt experience of producing the wrong thing herself.
The second bucket is simulated production: making juniors solve cases with the AI turned off, or with the AI’s output deliberately corrupted, so that they retain the ability to produce. Some hospitals are experimenting with versions of this in radiology training; some firms are experimenting with it in moot-court formats. The early results are interesting and, importantly, partial. They suggest that simulation can keep some skills alive but cannot fully reproduce the emergent judgment that used to come from doing the actual work, with actual stakes, repeatedly. Simulators teach pilots how to fly. They do not teach pilots how to be the kind of person who knows, on instinct, when something is about to go wrong.
The third bucket, and the most candid, is waiting. Many firms have decided, implicitly or explicitly, that the current cohort of seniors is sufficient for the next decade and that the question of generational reproduction can be addressed later. This is an institutional bet that the problem will resolve itself — that better tools, better training programs, or better juniors will materialize. It is not a strategy. It is the absence of one.
We are not in a position, in 2026, to say which of these will work, because none has been running long enough to find out. What we can say with some confidence is that none of them is, on its own, sufficient — and that the cost of getting this wrong is not borne in the year the bet is placed. It is borne fifteen years later, when the cohort that should have become senior partners and senior physicians and senior engineers shows up underweight on the very judgment those titles imply.
That, by itself, would be enough to call this the most important question in the essay. But the training mechanism is, in fact, the smaller of the two questions a profession-as-institution faces under these conditions.
The larger question is what a profession is, when its reproduction mechanism breaks.
Sociologists of work have a long literature on this.[4] A profession, in its strong sense, is not a job category. It is an institution that controls the production of its own producers — that decides who gets credentialed, what knowledge counts, what conduct gets a member expelled, and what the public is owed in exchange for that closure. The labor market is downstream of all of that. The labor market is what you see; the institution is what makes the labor market possible. When the institution’s reproduction mechanism breaks, the labor market does not immediately collapse — the existing seniors continue practicing, the existing credentialing bodies continue credentialing — but the institution slowly loses the thing that justified its claim to special standing in the first place. It loses the line that separated the people who do this from everyone else.
That line is being attacked from below as well. The writer Maggie Appleton, drawing on Mao-era Chinese rural medicine, has used the term barefoot developers to describe a coming class of end users who, equipped with AI tools, will produce the software they need without becoming software engineers in the institutional sense.[5] The frame extends naturally to other professions. The barefoot lawyer drafts her own simple contracts. The barefoot accountant runs her own books. The barefoot physician — and this is where the frame strains for obvious safety reasons — handles her own routine cases via tools that increasingly perform the differential diagnosis that used to require an office visit. None of these barefoot practitioners is replacing the senior professional. All of them are, however, redrawing the edge of the profession’s claim to its labor market.
What is happening, in other words, is a double erosion. The credentialed pipeline is unable to reproduce its seniors as fast as the seniors retire. The lay public is, simultaneously, encroaching on the simpler end of the work. The profession is being thinned in the middle and frayed at the edges at the same time. A field can survive one of these. We are not certain any field can survive both, in their current institutional form, at the speed at which they are now happening.
This is the question we want to spend the next section on. Not “will the professions adapt?” — every institution under pressure adapts, in the trivial sense that something will exist where they used to be. The question is whether what comes out the other end is recognizable as a profession at all, or whether the institution is replaced by something else: a regulated infrastructure, a credentialed thin layer of accountability over a substrate of barefoot practice, a guild reduced to the high-stakes cases. The answer matters because the institution is what produced the people we have been writing about, including the partner from section one and the radiologist from section three, and what replaces it will produce — or fail to produce — the people who will be doing this work in 2050.
The partner, by now, has gone home. Tomorrow he will go back to his office; on Friday he will sign off on a deal that the newest hire will have largely drafted; in fifteen years he will retire. The question is who will be in his chair.
Friday will look ordinary. The newest hire will have produced two more contracts of comparable quality to the first. The partner will read them, will catch a thing on page eleven that nobody else would have caught, will pick up the phone and explain to a CEO in language that does not appear in any draft why a particular clause needs to be the way the partner wants it. The deal will close. The invoice will go out. The hire will go home pleased with herself, and rightly so — she is good at this, she will get better, she will stay at the firm. Nothing on Friday will indicate, to anyone watching the firm from the outside, that anything has changed.
Everything has changed. The partner has spent his afternoon doing the only part of his job that has not yet been disassembled — the part where he knew, in ten seconds, the thing that mattered, and the part where he picked up the phone. That part is small. It is also what the firm is, increasingly, billing for. The hire has spent her afternoon doing the part of her job that the previous generation of partners did at her age, except that the part she is doing is no longer the part that was forging them. She is producing more, and learning less, and she does not know yet that this is the trade.
Somewhere across town, a radiologist of comparable seniority is having an analogous Friday. She has read forty-one studies before lunch. The tool flagged thirty-eight of them correctly. She caught the thirty-ninth herself. She does not know, and cannot easily find out, whether she would still catch the thirty-ninth in five years if her practice continued as it is going. She is still the radiologist of record, and the readings are still hers, and the patients she sees benefit from a system that is, on the whole, more accurate than the system that preceded it. None of that resolves the smaller, slower question.
A senior engineer at a hospital that depends on her work is, at this moment, approving a code change her tooling produced, that she has reviewed for ninety seconds. The change will ship before the weekend. Most weeks, the system she stewards absorbs the change without incident, and patients are scheduled, and films arrive at the radiologist on time. The week it absorbs a change it should not have is the week we will read about in the news, and not, yet, the week we are reading about today.
The thing the partner went home with — the thing the radiologist is taking home tonight, and the engineer is taking home tomorrow — is not, exactly, fear. It is older than fear, and quieter, and harder to act on. It is the recognition that the work to which they gave their adult lives turns out to have been two things welded together, and that the welding has come undone, and that the smaller of the two things — the part they actually built their lives around, the part that justified the long hours and the missed weekends and the children’s recitals attended late — is now visible for what it always was. Some of them will look at it and feel quietly proud. Some of them will look at it and feel exposed. All of them will eventually have to decide what they are going to do about the part they no longer get to do every day, that used to be how they kept the rest of it sharp.
The institutions they have built will decide, over the next fifteen years, whether the answer is found at the level of the firm, the field, or the public. The training mechanism is broken in a way that no firm has yet repaired. The boundary of the profession is being redrawn from below in a way no professional society has yet contested coherently. The cohort that should be replacing today’s seniors is, for the moment, mostly auditing instead of producing, and the difference will not show up on any quarterly result. It will show up the year a transaction the partner would have caught goes through, and is read about in the news, and a different set of partners begin to understand, for the first time, what was actually being lost.
Friday will look ordinary. Most Fridays, for the next ten years, will look ordinary. The day someone notices that the chair has been empty, in the sense that matters, for some time, is the day this essay will look like it was written about a problem that had a name.
Someone will be in the chair in 2040. We do not, in 2026, know what they will know.
For the AI-assistive deskilling literature, see esp. Artificial intelligence in medicine: a scoping review of the risk of deskilling and loss of expertise among physicians (PMC, 2025); also Artificial intelligence in clinical decision support and the prediction of adverse events (Frontiers in Digital Health, 2025) and Deskilling dilemma: brain over automation (PMC, 2025). The body’s claim represents the literature’s central, emerging finding. ↩︎
JPMorgan’s restriction on employee ChatGPT use was reported on 22 February 2023 by Bloomberg (https://www.bloomberg.com/news/articles/2023-02-22/jpmorgan-clamps-down-on-staff-s-use-of-ai-powered-chatgpt-bot) and contemporaneously by CNN (https://www.cnn.com/2023/02/22/tech/jpmorgan-chatgpt-employees/index.html). Samsung’s ban followed on 2 May 2023, after an April 2023 incident in which engineers reportedly uploaded proprietary source code to the tool; Bloomberg, https://www.bloomberg.com/news/articles/2023-05-02/samsung-bans-chatgpt-and-other-generative-ai-use-by-staff-after-leak; TechCrunch, https://techcrunch.com/2023/05/02/samsung-bans-use-of-generative-ai-tools-like-chatgpt-after-april-internal-data-leak/. JPMorgan’s internal LLM Suite was rolled out beginning July 2024 (Fortune, https://fortune.com/2024/07/26/jpmorgan-chatbot-llm-suite-chatgpt-ai-machine-learning-wall-street-banking-finance/) and expanded to roughly 140,000 employees by September 2024 per President Daniel Pinto’s remarks at the Barclays Global Financial Services conference (CIO Dive, https://www.ciodive.com/news/JPMorgan-Chase-LLM-Suite-generative-ai-employee-tool/726772/). ↩︎
Klarna’s original AI-assistant claims of 700-agent equivalent productivity and approximately $40 million in profit impact were announced in late February 2024; FinTech Magazine, https://fintechmagazine.com/news/klarna-ceo-ai-will-cut-workforce-by-a-third. The reversal and rehiring announcement came in May 2025; the operative quoted phrase (“a too predominant evaluation factor”) was given by CEO Sebastian Siemiatkowski to Bloomberg, with the full sentence reading: “As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.” The reversal was widely covered; see Fortune, https://fortune.com/2025/05/09/klarna-ai-humans-return-on-investment/, and Fast Company, https://www.fastcompany.com/91468582/klarna-tried-to-replace-its-workforce-with-ai. ↩︎
The locus classicus is Andrew Abbott, The System of Professions: An Essay on the Division of Expert Labor (University of Chicago Press, 1988), which argues that professions are best understood as overlapping jurisdictions over particular kinds of work, contested and renegotiated as conditions change. The framework is what allows us to say that a profession is more than its labor market — it is the institution that controls who is allowed to do the work, under what accountability, and on what claim to special knowledge. ↩︎
Maggie Appleton, Home-Cooked Software and Barefoot Developers, https://maggieappleton.com/home-cooked-software. The concept draws on the barefoot doctor program of mid-20th-century rural China, in which laypeople were trained in basic medical care to extend healthcare into communities the formal profession could not reach. Appleton applies the analogy to AI-assisted software development; we extend it, in this section, to the broader class of professional services. ↩︎
§