Microsoft Research's latest AI insights: The concept of "jobs" will become a thing of the past, and companies' organization, performance, and HR systems will face changes

Recently, Microsoft Research released a significant report titled "Working with AI: Measuring the Occupational Implications of Generative AI," which analyzed over 200,000 real conversations from Bing Copilot. The system analyzed how AI has helped in work, how well it performs, and which professions are most affected.

This report is not based on imagination or simulated experiments, but rather identifies patterns directly from real records of how people use AI every day, making it the closest research on AI's occupational impact to the current workplace reality.

The conclusion is direct and poignant: AI has quietly participated in many "seemingly safe" white-collar jobs, from writing copy, researching information to explaining complex content, many tasks no longer require human completion. Particularly, positions that originally relied on education, experience, and communication skills are being fragmented into tasks, with parts being taken over by Copilot. The research also found that higher-paying and higher-educated professions are more likely to be intervened by AI.

The report presents a core viewpoint: AI will not suddenly "replace a person," but is quietly "rewriting the definition of work." Future work will not be about sticking to one position, but rather a set of processes composed of many task modules. Those who can integrate AI into these tasks will significantly enhance efficiency. People who do not know how to use AI may not immediately lose their jobs, but will find themselves at a growing disadvantage in the workplace.

▍The Three Most Common Ways AI Helps

Based on 200,000 real user conversations from Bing Copilot, this report analyzed what tasks AI is actually used for in real work scenarios, how well it performs, and the extent of its impact on professions.

Overall, the tasks users most frequently ask AI for help with are: finding information, writing, and assisting in expressing opinions or explaining content. AI itself most commonly provides information, offers suggestions, and explains concepts, acting like an around-the-clock "knowledge assistant."

The research team referenced the U.S. Department of Labor's occupational database and categorized the tasks mentioned in the conversations into a system called "Intermediate Work Activities" (IWA). The results showed that the tasks AI is most commonly used to assist with are primarily concentrated in the following areas:

The first category is "finding information," such as researching data, checking product information, reviewing documents, and keeping professional knowledge updated. Copilot performs exceptionally well in these areas, not only appearing frequently but also receiving positive user feedback, with a like rate exceeding 80%. For example, when someone asks how to interpret a certain legal clause or what the principle of a certain medical condition is, AI can provide a clearly structured explanation within seconds, far surpassing the experience of traditional search engines The second category is "writing," which includes writing copy, articles, revising drafts, organizing materials, and so on. This type of task is not only the most popular but also best reflects the advantages of AI. Copilot can understand what the user wants to write, the tone they are using, and who the audience is, then help generate drafts or polish revisions with high efficiency. Many people use it to write resumes, speeches, blogs, work summaries, and more.

The third category is "clarifying," which means explaining some professional content to others. This includes many sub-tasks, such as explaining technical details, outlining policy processes, answering questions, and providing suggestions. Copilot's performance in this area can be likened to a tireless teaching assistant or customer service representative, capable of answering repeatedly, explaining clearly, and expressing ideas in an organized manner.

Additionally, research has found an important point: What users want AI to do does not always align with what AI actually does. In 40% of conversations, the user's goals and AI's actions are completely different; in over 90% of conversations, the overlap between the two is less than half. This indicates that, many times, AI does not directly replace users in completing tasks but rather acts as a "support," helping users better achieve their goals, such as organizing thoughts, supplementing information, translating professional language, and so on.

Among all these tasks, some of the most frequently occurring and effective ones include: editing documents, searching for information from various sources, providing explanations to clients or the public, answering questions, and preparing teaching or explanatory materials.

The Microsoft team also discovered something noteworthy: The high overlap between AI capabilities and tasks is driving a shift in job structures from "one person doing everything" to "humans responsible for judgment and creativity, while AI handles execution and expression." For example, in the job of a news editor, a person used to have to conduct interviews, write articles, and polish them. Now, while interviewing and topic selection still require human involvement, the initial draft writing and language polishing can already be handled by AI.

For instance, market researchers used to have to gather data, write reports, and analyze trends themselves; now, AI can scrape web pages, read industry reports, and generate drafts, with humans only needing to judge which sources are reliable and which logic holds up. This indicates that AI does not necessarily replace entire professions but rather "disassembles" jobs, taking away tasks that can be done piece by piece.

The Microsoft research team also emphasized one point — AI is best suited for tasks that can be clearly articulated and structured. As long as a task can be "clearly stated," AI can "perform adequately." However, as soon as a task involves judgment, interpersonal interaction, improvisation, or hands-on execution, AI is still far from being able to replace humans.

▍Who does AI assist? Who will it replace?

After analyzing the most common tasks performed by AI, the Microsoft Research Institute further investigated a key question: Which individuals are most likely to be assisted by AI, or even partially replaced? Are they high-salaried or low-salaried? Are they highly educated or not well-educated? They conducted a cross-analysis of the AI applicability scores for each profession along with information such as income, education level, and job nature (e.g., full-time or part-time), and the results were somewhat surprising.

First, let's look at income. The higher the salary, the greater the likelihood of AI involvement. In other words, for jobs in the top 25% of income, AI can not only be utilized but also performs quite well. Why is this the case? It's actually quite simple; many of these high-paying positions rely on cognitive skills, writing, analysis, and expression to earn money, such as economic consultants, policy analysts, market planners, technology journalists, and business editors. These are precisely the areas where Copilot excels.

For example, previously, if you needed to write a market trend analysis report, you would spend several hours reviewing reports, checking data, and organizing your thoughts. Now, Copilot can generate a draft in just a few sentences, and with some modifications, it can be ready to use, saving a significant amount of time.

Conversely, for low-paying jobs, such as cleaners, kitchen staff, movers, and couriers, the assistance AI can provide is actually quite limited. These jobs are not unimportant; rather, AI cannot step in. They do not involve typing reports or tasks that can be handled with a simple prompt; instead, they require hands-on work.

Looking at education level, the trend is similar. The higher the educational requirement for a position, the more likely AI can play a role. Jobs that require a bachelor's degree or higher often involve writing, speaking, summarizing, and analyzing, which are also Copilot's strengths.

For instance, researchers, content editors, knowledge-based bloggers, and industry consultants deal with information daily, either integrating viewpoints or producing content. Copilot can not only search for information but also help organize thoughts, polish language, and draft documents, making it a versatile assistant.

However, for positions with minimal educational requirements, such as warehouse stock clerks, laundry workers, kitchen assistants, and farm workers, AI finds it difficult to intervene. Asking AI to cook, move boxes, or organize goods is simply not feasible.

Thus, the conclusion drawn by Microsoft is quite interesting—it is not the "lower-level jobs" that will be replaced first, but rather "office workers" who will need to learn to work alongside AI.

In addition to income and education, the report also examined job nature, such as whether the position is full-time or long-term. The conclusion is also clear: AI is more likely to enter full-time, clearly defined, and highly repetitive positions. Many clerical jobs involve similar tasks daily, such as writing summaries, sending emails, researching, and preparing reports, where AI can be of great assistance. In contrast, tasks that are too fragmented or flexible, such as those of hourly workers, temporary promoters, or warehouse movers, make it difficult for AI to integrate.

The report also provided an interesting example: some jobs may have few tasks, but if those tasks align with what AI excels at, they will be significantly impacted. For instance, proofreaders and editors focus on correcting typos, polishing sentences, and standardizing formats; these tasks can be completed quickly and accurately by Copilot, potentially allowing AI to accomplish what a person would do in a day in just half an hour In other words, it is not that tasks are easily replaced by AI, but rather that “the tasks AI excels at are numerous,” making this position at risk.

The report also outlines the profile of the group most likely to be assisted by AI: generally, they have at least a bachelor's degree and engage in content-related, analytical, or communicative work, such as writing proposals, conducting industry research, and drafting educational copy. They earn decent salaries, their work is not overly burdensome, but the tasks are highly repetitive and clearly structured, making them particularly suitable for AI like Copilot to step in.

In other words, this group of people who used to rely on their brains and pens to complete their work now faces the reality that with AI's arrival, they must learn how to use AI as a partner, or else their efficiency will be outperformed.

The impact of AI is first altering the work methods of “middle-class white-collar workers,” rather than starting with jobs like assembly line workers or delivery personnel. Moreover, it raises the question not of “who will be laid off,” but rather “who needs to quickly learn how to use AI to offload some repetitive, expressive tasks.”

Regardless of how much you earn or your educational background, as long as your current job involves a lot of writing, speaking, or information integration, AI is likely already eyeing the part of your work that it can take over. The key to the future is not to evade AI, but to find ways to make AI your work assistant rather than a competitor.

▍What about “work”?

After analyzing what AI excels at and its significant impacts on certain individuals, Microsoft Research ultimately posed a larger question: What will AI ultimately transform “work” into?

In other words, AI is not just about helping you with a few tasks; it is gradually “rewriting” the definition of work. Previously, when someone was referred to as a “marketing manager,” “project assistant,” or “content editor,” a specific set of job responsibilities and rhythms would come to mind. However, now some tasks that originally belonged to these positions can already be performed by Copilot, and quite effectively.

Research has found that many people are no longer just using AI to “look up information” or “polish drafts,” but are actually allowing it to genuinely participate in their work processes. For example, they let it write first drafts, organize documents, prepare meeting content, and generate proposal frameworks—tasks that were previously done solely by humans, many of which have now been partially taken over by AI.

Microsoft's core viewpoint is: the concept of “positions” will gradually be replaced by “tasks.” A person will no longer be responsible for an entire job but will instead focus on some key tasks, while other tasks can be delegated to AI.

This change brings about a new question: how should companies allocate personnel and distribute work? Previously, hiring was done based on positions, with fixed tasks within those positions; but now, it is essential to clarify: which tasks can be handled by AI, and which still require human involvement. Organizational structure, performance evaluation, and HR systems must all adjust accordingly.

Some companies have begun experimenting with new division of labor: allowing Copilot to write emails, draft content, and organize client feedback, while humans handle emotional communication, strategic adjustments, and last-minute decision-making. As a result, team efficiency has actually increased, and client feedback has been positive This has also led to the emergence of some brand new positions, such as "AI Collaboration Manager" and "Workflow Designer," specifically focused on one thing: finding ways to make human and AI collaboration smoother and more efficient. They may not necessarily create content themselves, but they understand which tasks are suitable for AI and how to connect the front and back processes.

In this regard, Microsoft has summarized several keywords for the future of work transformation:

• Task Granularity: Work is not a position but a collection of tasks, breaking down which tasks AI can handle and which are better suited for humans.
• Human-AI Hybrid: It's not "AI does half, humans do half," but rather a close collaboration where you say one thing, and I say another, working closely together.
• Capability Restructuring: Humans handle creativity, judgment, and decision-making, while AI takes care of execution, expression, and organization.
• Role Fluidity: You do A today and B tomorrow; work boundaries widen, flowing according to tasks rather than sticking to a fixed position.

Overall, the key in the future is not whether you use AI, but whether you have the ability to "reorganize your work" and build a "human + AI" combination structure. Having the tools is one thing; how to use them efficiently is the dividing line.

For individuals, it's the same. Previously, learning new skills was about "being able to do more"; now the focus is on "which tasks can be handed over to AI," allowing you to free up time to engage in more valuable tasks that require human judgment. Those who know how to use AI may not necessarily win effortlessly, but those who do not know how to use AI will definitely become less competitive.

Microsoft also offered a particularly practical suggestion: In the future, what determines your workplace competitiveness is not how much you can do alone, but whether you have the ability to build a smart work combination that allows AI to assist you, fill in for you, and accelerate you. Those who can design this system will be more valuable.

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investing based on this is at your own risk