Don't be afraid of AI taking your job; learn how to coexist with it.
Written by: Unstable Paradigm
In February of 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 details thousands of professions and 19,530 types of job responsibilities. This data-driven matching clearly reveals how AI is integrating into various jobs and specifically which positions it is impacting.
(To protect privacy, the research team used a system called Clio, which only analyzes aggregated data and cannot access specific individuals' chat records.)
1. AI's Iron Fans: Not Bosses, but "Coders" and "Writers"
After the research results were released, the first finding was that AI usage is extremely "specialized." Nearly half of its applications are concentrated in two fields.
Champion: Computer and Mathematics (37.2%)
Indeed, AI's number one "iron fan" is the coder.
Imagine this scenario: Programmer Xiao Zhang is developing an e-commerce app when suddenly the program crashes, and the error message is incomprehensible. In the past, he might have spent half a day scratching his already sparse hair, desperately searching for the problem in a sea of code. Now, he throws the code and error message to Claude: "Bro, what's the problem?" AI quickly replies, "The issue is on line XX; this parameter format is incorrect."
From "developing and maintaining software" to "programming and debugging" to "designing databases," these are the tasks programmers most often ask AI to assist with. For them, AI is not a job thief but more like a tireless coding partner available 24/7.
Runner-up: Arts and Media (10.3%)
In second place are those who earn a living with their "writing skills." This seemingly "humanities" field actually collaborates exceptionally well with AI.
For example, Xiao Li from the marketing department needs to write a product promotion copy. She can have AI "brainstorm" several titles first, then choose the best one to continue writing. After finishing the first draft, she shows the article to AI: "Help me check if the language is engaging enough. Can it be more lively?" When she needs to publish the article in a specific format, AI can also quickly handle the layout.
These users include technical document writers, advertising copywriters, editors, and even archivists. For them, AI is the perfect combination of an inspiration repository, proofreader, and layout tool.
However, the distribution of AI usage across professions is severely imbalanced. As shown in the image below, computer and mathematics occupations, which account for only 3.4% of all jobs in the U.S., represent a staggering 37.2% of AI conversations; in contrast, food, sales, and transportation jobs, which collectively account for nearly 30% of the U.S. workforce, only make up 3% of the conversations.
Original image from Anthropic research dataset, this image generated using AI translation tools
2. Is AI a "Substitutor" or an "Enhancer"? Currently More Like a "Super Assistant"
After clarifying "who is using it," the next key question is "how is it used." The report provides an important statistic: 57% of usage falls under "enhancement," while 43% is categorized as "automation."
This indicates that currently, AI is more often acting as an "enhancer." Researchers categorized human-AI collaboration into five modes:
Automated Actions (43%)
Directive: The simplest form of "automation," akin to using a tool. "Translate this paragraph into English," and AI provides the result with almost no interaction.
Feedback Loop: Commonly used by programmers. The user provides code to AI, and after running into an error, they feed the new situation back to AI, looping until resolved. The human acts mainly as a "messenger."
Enhanced Actions (57%)
Task Iteration: Deep collaboration. You ask AI to design a webpage, and after AI provides the initial version, you say, "The layout is good, but the colors are too dark; can we brighten it up? Also, make the buttons larger." It's like two colleagues iterating together to complete a task.
Learning: Not to complete a task, but to gain knowledge. "Can you explain what a 'neural network' is using simple metaphors?" Here, AI acts as a versatile teacher.
Verification: You have completed a task but want AI to check it. For example, after writing SQL code, you ask AI to see if the logic is correct and if there are better ways to write it.
This 57% to 43% ratio indicates that most of the time, we are not passively "being served by AI," but actively "driving" AI. It acts more like a powerful external brain that we use to learn, iterate, and verify our work, ultimately making ourselves stronger.
3. Higher Income, More AI Usage? The Answer is "Inverted U-Shaped"
This may be a counterintuitive finding. The relationship between AI usage and salary does not rise linearly but follows an "inverted U-shaped" curve.
Both the bottom and top of the pyramid use it less.
Low-income jobs: Restaurant servers, construction workers, truck drivers. Their work requires a lot of physical labor and real-world interaction. AI currently lacks the ability to physically engage, making it difficult to participate.
Extremely high-income jobs: Surgeons, judges, senior management. These roles require not only top-level expertise but also carry significant responsibility and risk, with complex and uncertain decision-making processes. AI is currently far from reaching this level, and there are many legal and ethical restrictions.
Middle to high-income "technical white-collar" workers are the absolute main force. AI usage peaks in professions that require "a lot of preparation" but have not yet reached "top expert" levels, such as software developers, data analysts, financial analysts, and marketing managers.
This "inverted U-shaped" distribution clearly shows the current boundaries of AI capabilities. It excels at handling knowledge-based work that has fixed rules, is information and data-centric, but still requires considerable intellectual input.
4. AI is Blurring Professional Boundaries, Causing "Skill Inflation"
An interesting finding in the research is that many AI conversations classified as specific occupational tasks actually come from non-specialists. For example, queries categorized as "nutritionist work" may come from ordinary people seeking dietary advice rather than professional nutritionists.
This represents a new trend: AI is blurring professional boundaries, allowing ordinary people to enter fields that previously required specialized training. This "professional knowledge democratization" may lead to broader knowledge acquisition and application but also raises questions about professional value and quality control. When AI enables everyone to become "half an expert," how will the boundaries and value of professional services be redefined?
This also reveals another important trend: AI is creating a new "skill inflation." When AI can easily handle basic programming, "knowing how to code" is no longer an advantage. This will profoundly impact the job market and even society's definition of work. The definition of work itself has always been changing; decades ago, if you said "typing," people understood you were doing a specialized job. But now, if you say "typing," everyone thinks you are stating the obvious because typing is no longer seen as a professional skill, and thus the implied meaning of "working" in the phrase "typing" has disappeared.
As AI develops, many skills we consider valuable today may undergo similar transformations.
Conclusion: Don't be afraid of AI taking your job, but learn how to coexist with it
This report from 4 million real conversations paints a more complex and interesting picture than the "unemployment theory."
Overall, the AI revolution does not suddenly eliminate a profession but is a "penetration war," quietly changing every aspect of our work on a "task" basis. Research shows that about 36% of professions have at least a quarter of their job tasks affected by AI. In 4% of professions, the AI penetration rate of job tasks has exceeded 75%. Although this proportion is still not high overall, considering that we are only at the beginning of the AI era, this rate of penetration is already remarkable.
This penetration is silent and occurs even in fields that seem unrelated to technology. For example, lawyers may not be completely replaced by AI, but those who do not use AI for case research and document preparation may be surpassed by their peers who effectively utilize AI.
For each of us ordinary people, the biggest takeaway from this report is that at least in the short term, rather than worrying about being replaced by AI itself, we should be more concerned about being outperformed by those who are better at using AI.
The path forward thus becomes clear:
In the short term, learn to collaborate with AI, treating it as a highly capable co-pilot and tireless intern, helping us automate repetitive tasks, iterate creative work, verify ideas, and learn new knowledge.
In the medium term, learn to be the "boss" of AI. This requires skills: understanding AI's capability boundaries, precisely defining problems, breaking down tasks, issuing instructions, evaluating and integrating results, and leading workflows. This is not simple and requires skill and extensive practice.
Historically, every wave of technological advancement has followed the pattern of "eliminating old jobs and creating new industries." The steam engine eliminated the need for coachmen but created vast industries in manufacturing and logistics; electricity rendered lamplighters obsolete but ushered in a new era of electrical appliances and entertainment.
In the long term, AI will replace repetitive cognitive work, but this will not diminish human value; rather, it will make us more precious. We will no longer just execute tasks but ask questions; not just process existing data but bravely explore the unknown; not satisfied with imitation but pursue original ideas; not rely on cold interactions but build genuine connections with warm empathy; ultimately, what we seek is not efficiency but meaning. These are the heights of humanity that algorithms cannot reach.
You need not worry about AI, but rather about your own inability to use AI.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
