The narrative that AI makes human workers cheaper is inverted. In sectors with real emotional content — small services, hospitality, care — AI drives up the price of the human parts, because supply of humans who do them well is limited while demand rises. Klarna hired humans back in 2025. Duolingo learned the same lesson.
The narrative that AI makes human workers cheaper is upside-down. Klarna hired the humans back. Duolingo learned the same lesson. This essay makes the case that AI is not the enemy of human work — it is the mechanism by which certain kinds of human work will become worth many times more in the next decade. Written for small-business owners deciding whether to fire or restructure.
The dominant story about artificial intelligence and work, told at conferences and on podcasts since 2023, is that AI makes human labor cheaper. Employers, the story goes, will need fewer people. Wages for routine work will compress. The clever founders replace headcount with a subscription to a large language model and go home with the savings.
The story is wrong. It is not wrong about the direction of change — some kinds of work will be displaced. It is wrong about the price tag on what remains. The correct claim is inverted: in an economy where machines do most of the routine cognitive labor, the specific kinds of work that only humans can do become more scarce, more visible, and dramatically more valuable per hour.
The next decade is not going to be the decade of the disposable worker. It is going to be the decade of the worker whose presence, judgment, warmth, and character are what customers actively pay a premium for — precisely because those things cannot be produced by any system that runs on a graphics card.
This is a claim with commercial consequences, and small-business owners are the ones who will feel the consequences first.
Two well-documented examples from 2024 and 2025 show the reversal in real time.
Klarna — the Swedish payments company — announced in early 2024 that its AI customer-service assistant was doing the work of hundreds of full-time agents. Media coverage framed it as the future of white-collar work. In mid-2025, Klarna quietly reversed. Its own CEO, Sebastian Siemiatkowski, publicly acknowledged that the AI-only support strategy had degraded customer experience enough that the company had to rehire humans. The specific language used in his public comments was that cost-cutting had pushed too far and customers had noticed.
Duolingo — the language-learning company — announced in 2024 an AI-first restructure that laid off contractors. The Reddit and TikTok backlash was severe enough that the company's founder had to publicly clarify the policy within days. Duolingo continued using AI internally, but the public messaging around firing humans in favor of it was pulled back.
The pattern is not that AI failed. AI did what AI does — it produced acceptable-quality output for routine cases at very low marginal cost. The pattern is that customers noticed the human absence, and the absence carried a brand cost that exceeded the wage savings.
This is the mechanism. AI is not free. Its true price includes the value of the human presence it removes. Whenever the removed human was doing something the customer actually valued — recognition, judgment in edge cases, empathy in complaints, discretion, warmth — the removal shows up in the P&L on a different line, months later, as churn and NPS decline. The businesses that measured the wage savings without measuring the presence cost concluded that AI was a win. The businesses that measured both — like Klarna eventually did — concluded otherwise.
The counterintuitive mechanism works like this. When AI drives the cost of routine cognitive tasks toward zero, the price of the tasks that remain human is set by supply-and-demand for the residual work. The demand for the residual work does not fall — in most industries it rises, because more transactions happen, more customers exist, more edge cases occur. The supply of the residual work is constrained by the number of humans who are actually good at it.
Because supply is constrained and demand is stable or growing, the price per hour rises. This is basic labor economics applied to a bifurcating market. The routine cognitive worker whose job is fully absorbed by AI is displaced downward or out. The human worker whose job is the residual — the parts AI cannot do — becomes more expensive per hour than the same job was before AI existed.
Erik Brynjolfsson and colleagues at Stanford have published research on customer-service agents equipped with AI tools showing that the highest-performing agents get significantly more productive, and their value to the employer rises accordingly. The lowest-performing agents get lifted toward average — but they do not close the gap. The economic returns to individual human skill in the AI-assisted context appear, in these studies, to widen rather than compress.
For small businesses, this has a specific implication that will feel unfamiliar. The receptionist who knows every regular customer's name, the barber who remembers what school your kid started, the corner-store owner who lets people run tabs during the month — these are not sentimental legacy costs to be optimized away. In a market where the routine version of their job is available for pennies per interaction from any chatbot, the human version becomes rarer, more distinctive, and — this is the part small-business owners have not fully absorbed — chargeable at a premium.
There is a specific list, and it matters. The tasks that only humans do well, and where AI performance remains structurally weak, cluster around a few categories:
Recognition of the specific person. Not customer segmentation — actual recognition. Knowing that this particular customer went through cancer treatment last year and does not want to be asked cheerfully how she is doing. This is not data; it is context that lives in a human brain and is not extractable into a CRM field.
Judgment in edge cases. The customer whose situation does not match any script. The complaint that requires the operator to break the rule to keep the relationship. AI can be instructed to break rules under specific conditions; it cannot decide when the rule needs to be broken for reasons that are not in the instructions.
Emotional labor that persuades because it is genuine. The barista who genuinely cares that your mother just passed. The nurse who is unhurried at 11 PM. The mechanic who says the honest thing when the honest thing is that you do not actually need the repair. AI can perform care; it cannot feel care, and customers over time detect the difference.
Physical presence in the specific place. The corner shop that is open when it is snowing because the owner walked there. The stylist who came in on her day off for a wedding client. The dentist who called back personally. These are moves that only a human embedded in a specific community can make, and their value is not falling — it is compounding as the alternative gets more automated.
Trust that is transferred through personal relationship. The financial advisor who is trusted because his father was trusted by your father. The doctor who is trusted because you have been going to her for twelve years. This kind of trust is a slow-built asset; AI does not build it, and AI systems that mimic it are increasingly being flagged by customers as manipulative.
One honest concession: AI is genuinely better at scale-repetitive work. If your business is answering the same five factual questions to two thousand customers a week, an AI system does it faster and more consistently than any team. This essay is not an argument against AI. It is an argument against the specific mistake of treating AI as a replacement for the categories above.
There is a strategic framing that small-business owners rarely hear at conferences, because the conferences are being sponsored by the AI-implementation vendors: the moment your competitors all switch to AI-first customer service is the moment your all-human customer service becomes a distinctive market position, not a legacy cost.
The pattern has run before in adjacent categories. When large industrial food producers dominated the American food system in the 1990s and 2000s, the small farm was a legacy cost. When Whole Foods, farmer's markets, and farm-to-table restaurants scaled, the small farm became a premium position — the same operation, the same product, worth substantially more per unit because the alternative became universal. Independent bookshops during the peak of Amazon looked like sentimental legacy costs. The ones that survived to 2025 are, in many cases, growing revenue at a rate that would have surprised the industry in 2010.
The same asymmetry appears to be forming in consumer service. As the median customer experience across categories becomes AI-mediated, indifferent, and legible as automated to the customer, the businesses that maintain human warmth become distinctive in a way that they were not distinctive when everyone had it. The bakery where the owner knows your name is not a legacy cost when every other bakery has a chatbot on WhatsApp; it is a premium moat.
This is a bet, not a certainty. It could be wrong. The counterfactual is a world where customers acclimate to fully automated service and stop noticing the absence of humans. That world exists in some categories — utility billing, airline seat selection, self-checkout at scale — and the human-absence is now invisible there. But for categories that touch the emotional life of the customer — health, personal grooming, food, care of children and elders, celebration of milestones — the acclimation appears to have limits. The US Surgeon General's 2023 advisory on the loneliness epidemic did not appear from nowhere; the market is bending toward a customer base that is starving for real contact.
The long bet: your prime as a small-business owner is when your competitors all restructure to AI-first. If you have kept the receptionist, the barista, the salesperson who genuinely remembers people, you inherit the next-cycle customer base. AI stays; roles restructure; but the front line stays human, and the front line is what customers are increasingly paying a premium for.
The final claim of the essay is the one that softens the philosophy into practice. AI is not the opposite of the human worker. AI is the tool that makes the human worker capable of doing more of what only they can do.
The correct question for a small-business owner in 2026 is not: how do I reduce my headcount using AI? The correct question is: what is my most-skilled human doing right now that a machine could do, so that the human can spend that hour on the customer relationship that will keep the business alive in 2030?
The receptionist who is currently spending three hours a day on appointment reminders and payment link resends is not spending three hours on greeting regulars and de-escalating complaints. Automate the reminders. Do not automate the greeting. The stylist who spends two hours a week juggling booking messages is not spending two hours on the personal follow-up that turns a first-time customer into a five-year customer. Automate the booking. Do not automate the follow-up.
This is what AI is genuinely good for, and it is what the humans in your business will thank you for — because the humans in your business, if they are the good ones, are also tired of doing the parts of their job that a machine should have been doing five years ago. Give them back the emotional and judgment work. Take away the repetitive work. The math for the business is that the humans get more expensive per hour but there are fewer wasted hours; the customer gets a warmer experience per interaction with fewer interactions needed; and the business becomes the kind of operation whose customers explain, when asked why they keep coming back, that it just feels different here.
That is not sentimentality. That is a P&L strategy for the small business in the AI era. AI is not the enemy. AI is the chisel. The sculpture is your business — and the sculpture is human.