Who Is Losing Jobs to AI? The Answer Is Revealed in 4 Million Chat Logs
TechFlow Selected TechFlow Selected
Who Is Losing Jobs to AI? The Answer Is Revealed in 4 Million Chat Logs
Don't fear AI taking your job—learn how to work with it.
Navigating Web3 tides with focused insightsBy Fan Shiwending
In February this year, Anthropic, the company behind the Claude model, conducted a unique "workplace field study." They analyzed over 4 million user conversations and matched them with the U.S. Department of Labor’s O*NET occupational database, which meticulously documents thousands of professions and 19,530 types of job responsibilities. This data-driven matching clearly revealed, for the first time, how AI is integrating into various jobs and specifically impacting certain roles.
(To protect privacy, the research team used a privacy-preserving system called Clio, which only allows analysis of aggregated data without access to individual chat records.)
1. AI's Biggest Fans: Not Managers, But Coders and Writers
The first finding from the study was striking: AI usage is extremely skewed. Nearly half of its applications are concentrated in just two fields.
Champion: Computer and Mathematical Occupations (37.2%)
Yes, AI’s top fan base is software developers.
Imagine this scenario: Developer Xiao Zhang is building an e-commerce app when suddenly the program crashes, spitting out error messages that look like hieroglyphics. In the past, he might spend half a day scratching his already thinning hair, searching through oceans of code to find the bug. Now, he simply copies the code and error message into Claude: "Hey buddy, what’s wrong here?" The AI instantly replies: "The issue is on line XX—the parameter format is incorrect."
From "developing and maintaining software" to "coding and debugging" and "designing databases," these are the tasks programmers most frequently ask AI to help with. For them, AI isn’t coming to take their jobs—it’s more like a tireless, always-on programming partner.
Runner-up: Arts and Media Occupations (10.3%)
In second place are those who make a living with words—people in what sounds like very "liberal arts" fields, yet who turn out to work exceptionally well with AI.
For example, Li from marketing needs to write a promotional article. She can first ask AI to brainstorm several headlines, pick her favorite, and then continue writing. After drafting, she sends it back: "Can you check if the tone is engaging enough? Can it be more lively?" When she needs to format the article for specific platforms, AI handles layout instantly.
This group includes technical writers, copywriters, editors, and even archivists. For them, AI acts as a perfect blend of idea generator, proofreader, and formatting assistant.
However, AI usage across occupations is highly unbalanced. As shown in the image below, computer and mathematical roles—which make up only 3.4% of all U.S. jobs—account for 37.2% of AI conversations. In contrast, food service, sales, and transportation jobs—collectively nearly 30% of American employment—make up only 3% of AI dialogues.
Original data from Anthropic research; this image generated via AI translation tool
2. Is AI a "Replacement" or an "Enhancer"? Right Now, It's More Like a Super Assistant
After identifying "who’s using AI," the next key question is "how they’re using it." The report provides a crucial insight: 57% of use cases fall under "enhancement," while 43% fall under "automation."
This indicates that currently, AI functions more as an enhancer than a replacement. Researchers categorized human-AI collaboration into five patterns:
Automation behaviors (43%)
-
Directive mode: The simplest form of automation—using AI like a tool. "Translate this into English," and AI returns the result immediately, with little interaction.
-
Feedback loop: Common among programmers. Users submit code, run it, feed errors back to AI, and repeat until resolved. Here, humans mainly act as messengers.
Enhancement behaviors (57%)
-
Task iteration: Deep collaboration. You ask AI to design a webpage. After receiving the first draft, you say: "Layout looks good, but make colors brighter and enlarge buttons." It’s like two colleagues iterating together toward a shared goal.
-
Learning: Not task-oriented, but knowledge-seeking. "Can you explain neural networks using a simple analogy?" Here, AI becomes an all-knowing teacher.
-
Verification: You’ve completed a task and want AI to double-check. For instance, after writing SQL code, you ask AI to review logic or suggest better approaches.
The fact that enhancement (57%) outweighs automation (43%) shows that most users aren't passively served by AI—they're actively steering it. AI acts more like a powerful external brain, helping us learn, refine ideas, and verify outcomes, ultimately making us stronger.
3. Do Higher-Income People Use AI More? The Answer Is a "Reverse-U Shape"
This may be a counterintuitive discovery: the relationship between AI usage and income isn’t linear—it follows a "reverse-U" curve.
Both the bottom and top of the pyramid use AI less
-
Low-income jobs: restaurant servers, construction workers, truck drivers. These roles require heavy physical labor and real-world interactions. Since AI doesn’t have hands or legs yet, it naturally struggles to participate.
-
Very high-income jobs: surgeons, judges, senior executives. These positions demand elite expertise, carry significant responsibility and risk, and involve complex, uncertain decision-making. Current AI is far from capable in such domains, and legal and ethical constraints remain substantial.
-
Mid-to-high-income "technical white-collar" workers are the main users. Peak AI adoption occurs in roles requiring substantial preparation but falling short of "top expert" status—such as software developers, data analysts, financial analysts, and marketing managers.
This reverse-U pattern clearly reflects current AI capabilities. It excels at information- and data-centric knowledge work involving fixed rules and considerable cognitive effort—but not extreme complexity or accountability.
4. AI Is Blurring Professional Boundaries and Causing "Skill Inflation"
An intriguing finding from the study is that many AI queries classified as belonging to specific professional tasks actually come from non-experts. For example, questions labeled as "nutritionist work" often originate from ordinary people seeking dietary advice—not certified nutritionists.
This signals a new trend: AI is blurring professional boundaries, enabling non-specialists to enter areas previously requiring formal training. This "democratization of expertise" could broaden access to knowledge and application—but also raises concerns about quality control and the value of professional credentials. When AI enables everyone to become "half an expert," how will we redefine the boundaries and worth of true professional services?
It also reveals another critical trend: AI is driving a new kind of "skill inflation." When basic coding becomes effortless with AI, merely "knowing how to code" ceases to be a competitive advantage. This will profoundly reshape labor markets—and even society’s definition of work itself. The concept of work has always evolved. Decades ago, saying “I’m typing” clearly meant doing a skilled job. Today, saying “I’m typing” sounds trivial—because typing is no longer seen as a specialized skill. Thus, the implicit meaning of “working” embedded in those three words has vanished.
As AI advances, many skills we now consider valuable may undergo similar transformations.
Conclusion: Don’t Fear AI Taking Your Job—Learn How to Work With It
This battle report drawn from 4 million real-world conversations paints a far more nuanced and fascinating picture than the simplistic “AI will steal all jobs” narrative.
Overall, the AI revolution isn’t about abruptly eliminating entire occupations—it’s a “war of infiltration,” quietly transforming our work one task at a time. The study shows that around 36% of occupations already have at least 25% of their tasks influenced by AI. For 4% of jobs, AI penetration exceeds 75%. While still a small share overall, given that we’re only at the dawn of the AI era, this pace is astonishing.
This infiltration is silent—even happening in fields seemingly unrelated to technology. For example, lawyers may not be fully replaced by AI, but those who don’t use AI for case research and document preparation may be outperformed by peers who do.
For everyday workers, the biggest takeaway is clear: In the short term, we should worry less about being replaced by AI itself, and more about being surpassed by colleagues who use AI better than we do.
The path forward therefore becomes evident:
In the short term, learn to collaborate with AI—treat it as a super-powered co-pilot or an indefatigable intern. Let it automate repetitive tasks, iterate on creative work, validate your thinking, and help you learn new knowledge.
In the medium term, learn to be AI’s “manager.” This requires skill: understanding AI’s limits, precisely defining problems, breaking down tasks, issuing clear instructions, evaluating outputs, and orchestrating workflows. It’s not easy—it demands practice and technique.
Historically, every technological wave follows the pattern of “eliminating old jobs while creating new industries.” Steam engines displaced coachmen but birthed massive industrial and logistics sectors. Electricity made lamplighters obsolete but launched entirely new eras of appliances and entertainment.
In the long run, AI will replace routine cognitive labor—but this won’t diminish human value. On the contrary, it will highlight what makes us irreplaceable. We’ll shift from mere execution to asking meaningful questions; from processing existing data to boldly exploring the unknown; from imitation to pursuing original ideas; from cold interactions to building genuine connections through empathy. Ultimately, we’ll seek not just efficiency—but meaning. These are the heights of humanity that algorithms cannot scale.
You don’t need to fear AI—but you should fear the version of yourself who doesn’t know how to use it.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
